fix(runtime): typed-array own-property fallback shadows secret-key buffer metadata (#4363 regression)

[proggeramlug]

Summary

Main regression fix. #4363 (typed array own properties) added a typed-array own-property block at the top of js_object_get_field_by_name that, for a Uint8Array-backed buffer, return JSValue::undefined() for any key it doesn't recognize. A KeyObject's secret-key backing buffer is a Uint8Array, so its type / symmetricKeySize / asymmetricKey* metadata — resolved by the KeyObject block later in the same function — was shadowed and read back as undefined.

This is a deterministic failure on main HEAD (bisected to #4363): the unit test object::field_get_set::buffer_ic_miss_tests::secret_key_buffer_metadata_survives_ic_miss_for_aes_sizes fails, which fails the cargo-test gate on every open PR.

# on main (and #4363's parent passes):
test ...secret_key_buffer_metadata_survives_ic_miss_for_aes_sizes ... FAILED
panicked at field_get_set.rs: assertion `left == right` failed   (type read back as undefined, not "secret")

Fix

Fall through to the KeyObject metadata block for a secret-key buffer instead of returning the typed-array undefined fallback. Plain typed arrays are unchanged (still return undefined for unknown keys).

Validation

cargo test --release -p perry-runtime --lib: 979 passed, 0 failed (was 979 passed, 1 failed). The AES test passes both alone and in the full suite.

Found while diagnosing why an unrelated PR's cargo-test was red — the failure was this pre-existing main regression, not that PR.