gap suite red on main: typed array constructed with a fractional length reports .length === undefined (test_gap_3146, test_gap_4103) #7930

Found while running the gap suite for #7916 (PR #7928). Not caused by that branch — it
reproduces on a clean origin/main build with byte-identical output.

Symptom

A typed array constructed with a fractional length reports .length === undefined.

$ node --experimental-strip-types test_gap_3146_spec_throws.ts
new Int32Array(2.5).length = 2

$ ./perry-compiled
new Int32Array(2.5).length = undefined

ToIndex truncates 2.5 → 2, so the array is built with the right length; only the .length
read is wrong, which points at the length accessor / its fast path rather than the
constructor.

Two gap tests are currently red on main because of it

• test_gap_3146_spec_throws — new Int32Array(2.5).length
• test_gap_4103_typedarray_view_validation — u32 0: ok:undefined / u16 2,2: ok:undefined
  where node prints ok:2

Neither is in test-parity/gap_snapshot.json, i.e. the gate expects both to pass, so
scripts/run_gap_tests.sh is red on main.

Provenance (A/B'd, not assumed)

Built two compilers from origin/main @ ebefba51a and from the #7928 branch, ran both:

• cmp of the two outputs: identical on both tests.
• Reproduced with the default build and with PERRY_NO_AUTO_OPTIMIZE=1 + prebuilt
  runtime archives, so it is not a knob or link-mode artefact.

Parity is tag-gated (it does not run per-PR since v0.5.1018), which is presumably how this
landed and stayed.