fix(runtime): TypedArray lastIndexOf routes to array path, not string (#2457)

[proggeramlug]

Fixes #2457.

Problem

new Int32Array([1,2,3,2,1]).lastIndexOf(2) threw TypeError: (number).lastIndexOf is not a function, while .indexOf / .includes worked (fixed in #2456).

A --trace llvm showed the divergence: ta.indexOf → js_array_indexOf_jsvalue (array path), but ta.lastIndexOf → js_string_last_index_of (string path). The HIR lowers lastIndexOf on a known-not-string / typed-array local to the string method; at runtime the typed-array pointer reads as a number, so the string dispatch throws. indexOf/includes avoided this via dedicated Expr::ArrayIndexOf/ArrayIncludes lowering — lastIndexOf had no equivalent.

Fix

Add a parallel Expr::ArrayLastIndexOf { array, value, from_index }:

• HIR (local_array_methods): emit it for a known-not-string / typed-array local; add lastIndexOf to the ambiguous-method set so an unknown-typed local (possibly a string) still falls through to the string path.
• Codegen (misc_methods): lower to js_array_last_index_of_jsvalue (value + optional fromIndex with spec clamping), mirroring the lower_array_method::lastIndexOf arm.
• Runtime (array/search): re-add the typed-array branch to js_array_last_index_of_jsvalue (now reached) — backward strict-equality scan over the typed store via js_typed_array_get.
• Wire the new variant through the walkers, collectors, monomorph substitute, and stable-hash (exhaustive matches).

Uint8Array/Buffer keep their existing byte-level runtime dispatch (they skip the array block).

Testing

• cargo test -p perry-runtime --lib → 795 passed, 0 failed (adds typed_array_last_index_of).
• Byte-for-byte vs node, all match:
  • Int32Array/Float64Array lastIndexOf (incl. fromIndex and NaN),
  • regular-array, string, and any-typed-param lastIndexOf (unchanged — no mis-routing),
  • Uint8Array lastIndexOf (byte dispatch unchanged).

Follow-up to #2456 (TypedArray indexOf/includes); part of #2447.