opt-report v2: LocalId → Span side-table so --opt-report can print file:line and a source snippet #7036

Follow-up to #6952 (--opt-report, which shipped v1).

What v1 does

--opt-report reports each value by function + binding name:

  batch.ts :: local `acc` [loop depth 0]
      in function totalsRow -> Boxed
      ptr-shape rule 2 (containment)
      returned from this function. ...

The gap

HIR carries names through lowering but not source positions. Stmt::Let
is { id, name, ty, mutable, init } — there is nowhere to hang a span. So a
denied value cannot be rendered as file:line with a source snippet, which
is what makes an LLVM-style optimization remark actually pleasant to read.

One exception already exists and v1 uses it: Expr::New carries byte_offset
(#5253, captured for constructor TypeError locations), so unbound allocation
sites — the .map(x => ({...})) idiom, which has no binding name at all —
report a byte offset. Everything else has nothing.

Proposal

A LocalId -> Span side-table populated during AST→HIR lowering:

• SWC spans exist upstream at every Stmt::Let lowering site.
• perry-diagnostics (span.rs, source_cache.rs, emitter.rs) can already
  resolve a span to file:line:col and render a snippet with a caret.
• Strictly additive: v1's output gains a location line; nothing about the
  existing schema changes. Entry.byte_offset is already in the
  --opt-report=json schema (schema_version: 1) and would simply be
  populated for more positions.

Why it is its own PR

Threading spans through AST→HIR lowering touches every Stmt::Let construction
site across perry-hir, and the side-table has to survive the transform passes
(closure conversion, async lowering, inlining) that rewrite locals. That is a
scope-check of its own, not a rider on a reporting feature.

Acceptance

--opt-report on a program with a boxed local prints the declaration's
file:line:col and a source snippet, and the gap suite is unaffected.