fix(types): bind package-local bare type refs to their own module identity (#2208)

[gertybotbot]

Summary

Fixes #2208. A bare type reference inside a non-root module — e.g. a Holder { w: Wrap } field whose Wrap is declared in the same package-local module — resolved through the is_local arm of resolve_type_expr to the bare type_defs key. That key is last-write-wins across modules that declare the same bare name (pre_register_type_decl documents the qualified alias as the authority), so when a sibling module declares its own Wrap, the field silently binds to whichever module registered Wrap last.

Observability (correcting the issue note)

The issue notes this as "unobservable in correctly-disambiguated programs." It is in fact observable. Two package-local modules, each with a divergent Wrap and a Holder { w: Wrap }, mis-resolve one holder's field to the other module's layout:

error: no field `a_field` on type `Wrap`
  = help: b_field

…and the winner flips with import order (a program that reads a.w.a_field / b.w.b_field compiles or fails purely by which module's Wrap was registered last).

Fix

In the is_local arm, bind to the current module's own qualified identity ({short}.{name}) when that key exists — mirroring the sibling published_bare_type_qualified branch that already qualifies specifically to dodge this bare-key collapse. pre_register_type_decl always seeds the qualified key alongside the bare one, so:

• non-colliding locals resolve identically to before,
• root-module references (current_module_short() == None) are untouched,
• only the cross-module collision case changes — from wrong-module to own-module.

Boundary / neighbours checked

• Root program (current_module == None): unchanged, still binds bare.
• Single-module local reference: unchanged (qualified key == own def).
• The published_bare_type_qualified and handle_matches arms below are untouched.

Tests

New load-bearing regression same_bare_name_member_field_binds_to_own_module_identity asserts each holder's w field carries its own module's Wrap identity (alpha.Wrap / beta.Wrap). It fails on revert (both collapse to the single last-writer identity), verified locally.

• hew-types: 1665 lib + all integration suites green (registry_core 281→282).
• hew-hir: full suite green (field identity flows into HIR/MIR).
• cargo fmt + cargo clippy --tests: clean.
• e2e: the two-module repro (both import orders) now prints 42 / hello and exits 0.

[slepp]

Thank you for the #2208 report and the core own-qualified binding fix. I preserved your commit and authorship in #2626, which adds the private nested-module visibility fix and the full corpus proving needed to complete this change. Superseded by #2626.

[slepp]

Thanks for this — the #2208 collision fix is correct in intent and the observability correction is right. Review found one boundary hole: the own-qualified binding trips the visibility gate for same-module references to private types in nested modules (it broke std/net/tls), because the gate compares the short decl-module against the full accessor path. I've carried your fix forward in #2626 with that boundary closed and a regression covering the private-local-in-nested-module case (which this PR's pub-only test didn't exercise). Closing in favour of #2626, with credit for the original fix and the sharper issue diagnosis.