parse_schema: report when Formats/Latest was truncated at SCHEMA_SCAN_LIMIT

[DrunkOnJava]

Problem

rvt::formats::parse_schema (src/formats.rs, ~line 544) scans at most the first 64 KiB of the decompressed Formats/Latest stream:

const SCHEMA_SCAN_LIMIT: usize = 64 * 1024;
let data = if decompressed.len() > SCHEMA_SCAN_LIMIT { &decompressed[..SCHEMA_SCAN_LIMIT] } else { decompressed };

That cap is deliberate (the tail is binary object data that trips the class-name heuristic), but nothing records that it was hit. On real project files the stream is far larger than 64 KiB, and the Discussion #112 investigation (docs/disc-112-coordination.md) found this silent cap is the dominant boundary on how much of the schema we ever look at. Callers — rvt-schema, rvt-inspect, the formats_latest_integrity diagnostic added in #168 — cannot tell "the schema fit" from "we stopped looking".

What to do

1. Add a field to SchemaTable (e.g. scan_truncated: bool, or scanned_bytes + total_bytes) set by parse_schema when the cap applies. Parsing results must not change.
2. Surface it in at least one user-visible place: the Formats integrity row of rvt-inspect / export_diagnostics.formats_latest_integrity is the natural home (see src/compression.rs FormatsLatestIntegrity and how feat(diagnostics): Formats/Latest multipage integrity (A2/PARSE-001) #168 wired it through), or the rvt-schema summary line.
3. Unit test: a synthetic buffer of >64 KiB (a tiny valid schema prefix followed by zeros) sets the flag; a small buffer does not.

Acceptance

• tools/check-local.sh green; existing schema-classification tests unchanged.
• No claim that the tail is parseable — this is a "we stopped here" signal, not a fix for multi-page Formats/Latest (that stays tracked under RE: Reproduce checksum-paged stream decompression and silent corruption #151 / RE: Reproduce complete Formats/Latest parsing and serialization-tag assignment #154).

Scope notes

Small Rust change + test; runs on the checked-in tier-1 fixtures (they are under 64 KiB, so the flag is false for them — the >64 KiB case is the synthetic buffer in the unit test). Ask on the issue if the diagnostic plumbing is unclear.