test(self-hosting): add executable milestone validation

[graydeon]

Summary

The self-hosted compiler currently has structural smoke coverage, but not enough executable milestone coverage to make progress measurable with confidence.

There is also a known large-source validation gap: one of the self-hosted smoke tests is ignored because concatenating the self-hosted modules into one large source string triggers a parser/session failure path.

Why this matters

• self-hosting is a core project goal
• green test status can overstate maturity if the hardest self-hosted path is ignored
• milestone progress is easier to trust when behavior tests exist, not only parse/typecheck shape checks

Current behavior

• self-hosted smoke tests validate symbol presence and type-checking for staged concatenated modules
• a larger all-modules validation path is currently ignored because large concatenated input is not stable enough
• several self-hosted modules still expose placeholder behavior, so the test strategy needs to distinguish structural readiness from executable readiness

Expected behavior

• the self-hosted test suite should include milestone-oriented behavioral checks
• the large-source validation path should either work reliably or be replaced with a module-aware harness that exercises the same coverage without concatenation fragility

Suggested scope

• investigate and fix the large-source Session::from_source() / parser-state failure mode, or replace it with a better validation harness
• add milestone tests that verify actual self-hosted behavior for the earliest supported compiler stages
• keep structural smoke tests, but supplement them with behavior-oriented checks

Acceptance criteria

• no ignored test remains solely because the self-hosted source bundle is too large for the current session/parsing path
• at least one self-hosted milestone is validated with behavior assertions rather than only symbol/typecheck checks
• the test suite makes it clear which self-hosted stages are structural vs executable