fix: key resource-import fallback by name, eliminate sentinel (closes #99)

[avrabe]

Summary

Closes #99.

build_resource_type_to_import previously had a fallback branch that
collapsed every [resource-rep] / [resource-new] core import onto a
single map slot keyed (0u32, prefix) whenever a component imported
resources without canonical entries. That sentinel was intentional for
the single-resource case but is a known foot-gun for any future code
path that writes multiple resources into it.

Approach: option (b) — key by resource name

Per the issue's two options for hardening, this PR implements option
(b): key the import-only fallback by the resource short-name extracted
from the import field (e.g., [resource-rep]frequency → key
("frequency", "[resource-rep]")). The sentinel is gone, and so is the
downstream used_sentinel gate in resolve_resource_positions.

Why option (b) over (a):

• The fallback is genuinely useful (covers import-only components), so
  preserving it with safe keying is preferable to asserting a
  single-resource invariant that future work could legitimately need to
  break.
• The downstream consumer already relied on a sentinel check to mark
  resolved entries as "callee does not define this resource"; with the
  sentinel removed, that signal becomes "matched via the name index"
  rather than "matched via type-id 0", which is a cleaner predicate.

Concrete changes

• New ResourceImportMap struct with two indices:
  • by_type_id: HashMap<(u32, &'static str), (String, String)> —
    populated from canonical ResourceRep / ResourceNew entries plus
    Step-5 missing-op inference and Step-6 alias propagation. Identical
    semantics to the pre-existing primary map.
  • by_name: HashMap<(String, &'static str), (String, String)> —
    populated by scanning core module imports for [resource-rep]<rn>
    / [resource-new]<rn>. Replaces the sentinel slot.
• New ResourceImportMap::resolve(component, type_id, prefix) returns
  Option<(&entry, matched_via_name_fallback: bool)>. Try by_type_id
  first; on miss, translate the type id to a resource short-name via
  ParsedComponent::resolve_resource_type and look up in by_name.
• resolve_resource_positions — drops the used_sentinel gate. The
  flag now reads "did this match come from the name index?", which is
  the same logical condition without the sentinel.
• resolve_inner_resource_imports / resolve_one_layout — extended to
  take the parsed component so they can use the name fallback for
  inner-list resource borrows.
• All call sites in resolver.rs updated; no signatures outside the
  module changed.

Test plan

• cargo test --lib -p meld-core — 181 unit tests pass (was 177 +
  4 new for issue Design: resource-fallback sentinel in build_resource_type_to_import #99).
• cargo test --release --test wit_bindgen_runtime — 73 tests
  pass.
• cargo test --release — full workspace green.
• cargo fmt --all -- --check clean.
• cargo clippy --workspace --all-targets -- -D warnings clean.
• Pre-commit hooks (cargo-fmt + cargo-clippy + cargo-test) pass.

New tests

meld-core/src/resolver.rs:

• test_issue_99_multi_resource_import_only_no_collapse — hand-built
  ParsedComponent with two distinct [resource-rep] and two
  [resource-new] imports; asserts 4 distinct entries in by_name,
  and that x and y resolve to different module/field tuples.
• test_issue_99_single_resource_import_only_keyed_by_name —
  single-resource case keys by the actual resource name, not by an
  empty-string sentinel.
• test_issue_99_no_resources_produces_empty_map — empty fallback
  case is genuinely empty.
• test_issue_99_get_by_name_disambiguates_resources — direct
  get_by_name on two resources from different interfaces returns
  the correct per-resource entries.

🤖 Generated with Claude Code