feat(app-page): vendor tools as Nipype interfaces & Pydra tasks

[nx10]

What & why

@styx-api/core 0.5.0 (now latest) adds experimental nipype and pydra code generation. This surfaces both in the "Vendor this tool" (Source) tab so users can vendor a tool as a Nipype Interface or a Pydra task — not just a raw Python/TS wrapper — and reworks the picker, which had outgrown its flat tab strip.

The key wrinkle: 2-file delegation

generateNipype/generatePydra don't produce single-file drop-ins. Each emits an interface/task file that imports the styx Python wrapper (from ._bet import bet, BetOutputs). So those targets render two files — <iface>.py + _<styx>.py (the latter reusing the existing pythonModule) — with a "drop both files together" note. Filenames come from the compiler's own stems, never the raw tool name, which is required for the relative import to resolve (e.g. afni/3dcalc → from ._v_3dcalc import v_3dcalc, since 3dcalc can't name a Python module).

Changes

• Bump @styx-api/core 0.4.0 → 0.5.0.
• Worker emits generateNipype/generatePydra defensively (try/catch → null, with a console.warn so a genuine codegen regression is distinguishable from the intended per-tool "not applicable" skip), plus a canonical moduleStem from buildTypedSpec.
• Thread a DelegationArtifact type + moduleStem through the compile protocol → client → AppPage → ResultsPanel.
• UX: the cramped 5→7 tab strip becomes a grouped Select (Language wrappers / Workflow frameworks (Python) / Specs) + content panel. Nipype/Pydra carry an Experimental badge and a caveat note. The shown target is derived (falls back to the first available option) instead of reset in an effect.
• Consistency fix (pre-existing): all vendored filenames now use the compiler's module stem, so e.g. 3dcalc yields import-valid v_3dcalc.py / v_3dcalc.ts matching the delegation files, instead of the non-importable 3dcalc.py.

Heads-up

Bumping to 0.5.0 will trip the existing "Compiler version mismatch" banner until the published niwrap manifest is rebuilt with 0.5.0 — that's the lockstep warning working as intended, not a regression.

Verification

• npm run check (0 errors), npm run lint, npm run test (185 passing) all green.
• In-browser (fsl/bet + afni/3dcalc): grouped Select shows 3 headings / 7 options / Experimental badges; Nipype → bet.py + _bet.py with a resolving import + note; 3dcalc → v_3dcalc.py / _v_3dcalc.py; Python/TypeScript/Boutiques/schemas all still render; no console errors.