fix(codegen): closes #317 — wire String.matchAll codegen + NaN-box inner arrays (v0.5.415)

[proggeramlug]

Summary

• Adds the missing Expr::StringMatchAll codegen arm in crates/perry-codegen/src/expr.rs. Mirrors Expr::StringMatch exactly: SSO-safe receiver unbox via unbox_str_handle (same String-array element access: arr[i] SIGSEGVs and indexOf returns -1 #214 SSO bug class), regex handle via unbox_to_i64, call I64 js_string_match_all, then nanbox_pointer_inline on the resulting *mut ArrayHeader.
• Declares js_string_match_all (I64, I64) -> I64 in runtime_decls.rs immediately after the existing js_string_match decl.
• Fixes a latent bug in the existing crates/perry-runtime/src/regex.rs::js_string_match_all: outer-array slots holding the per-match inner array were being stored as f64::from_bits(inner_ptr as u64) (raw pointer bits). The codegen IndexGet fast paths (post-Array's hole should be undefined rather than 0 #323) expect NaN-boxed slot values, so m[1] / m[2] would have read back nonsense without this fix. Switched to js_nanbox_pointer(inner as i64) — same convention as js_array_group at array.rs:1493 and as the inner string elements.
• New regression test test-files/test_issue_317_string_matchall.ts covers: the issue's literal repro (for...of over /([a-z]+)=([a-z]+)/g with capture group access), three-capture-group dates with full-match m[0], no-match returning empty iterable (matchAll never returns null), spread [...text.matchAll(re)] with index-into-result, and single-match.

Surfaced compiling effect/src/internal/cause.ts (func 140) during the #309 compat sweep. Companion follow-up under umbrella #321.

Test plan

• cargo build --release -p perry-runtime -p perry-stdlib -p perry clean
• node --experimental-strip-types test-files/test_issue_317_string_matchall.ts matches Perry binary output byte-for-byte
• Gap tests: 27/28 = baseline (lone fail is pre-existing console_methods ci-env quirk)
• Parity: 179/179 (100%)

Closes #317.