feat: absorb 6 kernel-surface files from Modules/Core

[smuchow1962]

Summary

Modules/Core's test project cannot compile because 11 files exist only in
Core, not in Herald.OSS, and the test project references Herald.OSS (not
Core's own assembly) to avoid a version collision. This PR moves the 6 of
those 11 files that carry no licensing gate and no naming collision into
Herald.OSS, so Core's test project can pick them up through the reference it
already has.

• Adds FlightRecorderCommunityShape, PipelineCapabilityMatrix,
  PipelineEditionValidator, PipelineManifestReader (Configuration), and
  IStructuredOnlySink, MaterializingKernelSink (Pipeline/Kernel).
• Adds the 5 corresponding test files, trimmed and fixed to run standalone
  in an OSS-only test process (no Pro/Enterprise plugin loaded).
• Adds a new MaterializingKernelSinkTests.cs written against OSS's own
  SpyLogger test double, instead of importing Core's 636-line
  KernelPipelineTests.cs wholesale — that file mixes general kernel
  coverage OSS already tests under its own file layout with the 2
  MaterializingKernelSink tests this PR actually needs.
• Adds the 11 .mmpform dashboard config-form manifests these types read
  at runtime for OSS's own Community-tier pipeline steps.

Real defects surfaced / Follow-ups

• FlightRecorderCommunityShapeTests expected InvalidOperationException;
  OSS's HeraldEditionGate throws the more specific
  HeraldEditionRequirementException (not a subclass — direct
  Exception). Fixed the test to expect the type OSS actually throws.
• Manifest container headers used Core's internal/technical step names
  ("Post-Filtering (NLog PostFilteringWrapper equivalent)") while OSS's own
  PipelineStep definitions carry plain-language DisplayNames ("Filter on
  Message Text"). Prefixed each of the 10 Community-tier manifest headers
  with OSS's own display name so the dashboard label check passes and the
  copy matches what OSS actually calls the step.
• One assertion in Catalog_includes_every_matrix_entry was
  order-dependent. PipelineCapabilityMatrix.All caches a
  FrozenDictionary at first touch, while PipelineStep.AllNames (and so
  PipelineCatalog.Build()) stays live. Other test classes in this
  assembly (PipelineStepCapabilityTests, DetourCSurfaceFillInTests)
  already call PipelineStep.Register(...), and depending on run order
  those registrations can land after the matrix cache warms up. The
  assertion checked that the catalog and matrix counts matched; loosened it
  to the subset check the test actually needs (every matrix entry has a
  catalog entry) — that invariant holds regardless of ordering. This is a
  pre-existing characteristic of
  PipelineCapabilityMatrix's caching design, not something this PR
  changes; flagging for whoever owns that type next.
• Not moved: QuickLogBuilder.With.cs, SinkRegistry.cs, SinkInfo.cs,
  SinkNameValidator.cs, NetworkSinkNameAssigner.cs — the remaining 5 of
  the 11 Core-only files. Together they implement a "named sink" feature.
  Core's WithConsoleSink/WithFileSink/WithHttpJsonSink add a 3rd
  name parameter that OSS's existing Sinks.cs/NetworkSinks.cs
  overloads don't have — moving them as-is produces ambiguous-overload
  errors against OSS's 2-param overloads once a caller omits the optional
  tail. The feature also touches already-diverged shared files
  (QuickLogResult.cs: 856 diff lines against Core's copy,
  QuickLogBuilder.cs: 186, PipelineBuildResult.cs: 13,
  SinkRuntimeOverrideSet.cs: 170), and something in Core's own reference
  graph — not its plain OSS source tree — already supplies these members to
  Core's test project. That needs investigating before a real port; it
  isn't a mechanical file move. Filing as a follow-up rather than guessing.

Test plan

• dotnet build tests/Herald.OSS.Tests.csproj -c Release -f net8.0/net9.0/net10.0 — all three green, 0 errors, same 3 pre-existing warnings on each TFM as before this PR.
• dotnet test tests/Herald.OSS.Tests.csproj -c Release -f net10.0 — 1606 passed, 1 skipped, 0 failed (baseline: 1541 passed, 1 skipped).
• dotnet test tests/Herald.OSS.Tests.csproj -c Release -f net9.0 — 1606 passed, 1 skipped, 0 failed (baseline: 1541 passed, 1 skipped).
• dotnet test tests/Herald.OSS.Tests.csproj -c Release -f net8.0 — 1601 passed, 1 skipped, 0 failed (baseline: 1536 passed, 1 skipped — net8 omits 5 tests present on net9/net10, consistent with the baseline gap).
• Reran net10.0 and net9.0 three additional times each to rule out flakiness on the newly-added/edited tests specifically — all clean. One transient failure was observed once each on net8.0 and net9.0 during this work, in tests unrelated to this PR's changes; it did not reproduce on rerun and is consistent with this repo's already-tracked test-isolation-flakes note.
• Not covered: the 5 not-moved files above have no test coverage change here — they stay wherever Core's follow-up lands them.

Verified per TFM

Run	net8.0	net9.0	net10.0
Full suite, this PR	✅ 1601 passed, 0 failed	✅ 1606 passed, 0 failed	✅ 1606 passed, 0 failed
Baseline before this PR	1536 passed, 0 failed	1541 passed, 0 failed	1541 passed, 0 failed

Full test-suite results — net10.0

Passed!  - Failed:     0, Passed:  1606, Skipped:     1, Total:  1607, Duration: 18-33 s - Herald.OSS.Tests.dll (net10.0)

Full test-suite results — net9.0

Passed!  - Failed:     0, Passed:  1606, Skipped:     1, Total:  1607, Duration: 18-25 s - Herald.OSS.Tests.dll (net9.0)

Full test-suite results — net8.0

Passed!  - Failed:     0, Passed:  1601, Skipped:     1, Total:  1602, Duration: 17-18 s - Herald.OSS.Tests.dll (net8.0)

Addendum: blast radius

• Touches PipelineCatalog.cs (one comment edit only — no behavior
  change): the doc comment said OSS "intentionally omits" the matrix shape;
  it no longer does, so the comment now says the matrix lives here but
  PipelineCatalog still deliberately keeps its own hand-maintained cost
  numbers instead of consuming it.
• New embedded resources (manifests/pipeline/*.mmpform, 11 files):
  OSS's own Herald.OSS.csproj already had a conditional
  EmbeddedResource Include="manifests\pipeline\*.mmpform" wired in,
  unused until this PR populates the directory — this was an anticipated
  seam, not new plumbing.
• No existing public type's signature changes. Every added type is new;
  nothing already shipped in Herald.OSS moves or changes shape.
• Merge order: this PR merges before
  mmpworks/Herald.Core#refactor/drop-absorbed-kernel-surface (companion
  PR). Core's test project references Herald.OSS by sibling ProjectReference
  when present, falling back to the published Herald.OSS NuGet package
  (currently pinned at 0.12.11 in the umbrella's Directory.Packages.props)
  otherwise. Until this PR merges and a new Herald.OSS version ships, the
  Core PR's test-file removal has nothing to fall back on for a
  from-scratch clone that restores the package instead of building the
  sibling project. No HeraldCoreVersion or Directory.Packages.props
  pin needs to move for the PAIR to be internally consistent (both repos
  build via the sibling ProjectReference today, in this same umbrella
  checkout) — but the next Herald.OSS NuGet release should include this
  change before anyone rebuilds Modules/Core from a package-only
  restore.

---

AI-assisted: written with Claude (Max, build/licensing steward persona) via Claude Code, directed and reviewed by Steve.

[smuchow1962]

Companion PR (merges after this one): https://github.com/mmpworks/Herald.Core/pull/3