fix(extract): resolve alias re-exports and imports through barrels to the defining symbol (#1983)

[HerenderKumar]

Summary

Follow-up to #1983. e5a4b51 fixed the issue's direct reproducer, but its rewrite learns old→canonical symbol forms only from symbols a file defines. A barrel defines nothing — so a named re-export or import that resolves through a barrel still synthesizes an absolute-prefixed target no rewrite ever learns: the checkout path leaks into the edge target and the edge dangles.

Verified on current v8 head:

re_exports: src_barrel2 -> <checkout>_src_lib_index_formatdate  (dangling)
imports:    src_consumer -> <checkout>_src_lib_index_formatdate  (dangling)

for export { formatDate } from '@/lib' / import { formatDate } from '@/lib' where src/lib/index.ts is itself a barrel re-exporting from @/lib/utils.

Fix

Three parts, extending existing machinery rather than adding a new pass:

• _import_js stamps target_file on symbol-level re_exports/imports edges — the same transient hint the file-level edge has carried since Phantom self-cycles: same-basename cross-extension re-exports (X.ts → ./X.mjs) collapse to one node #1814 (no downstream reader; stripped at build).
• The id-remap post-pass gains a repoint step after the candidates rewrite: decompose a still-unowned target into (canonical file stem, symbol) via the stamp, follow the barrel's own already-canonical re_exports edge to the defining symbol — iterating one hop per pass so multi-hop barrel chains resolve — and, when no chain leads to a real node, canonicalize the prefix anyway so a checkout-specific path never survives in an edge target.
• The redundant-twin dedup now excludes target_file from edge identity, so an alias edge (stamped) still matches the canonical twin the shared resolver emits (unstamped).

Testing

• 3 new tests: single-hop barrel re-export resolves to the defining symbol; two-hop chain resolves; no re_exports/imports target contains the checkout prefix (the issue's core "Expected" line).
• Red-green verified: with the production change reverted, exactly the 3 new tests fail; restored, tests/test_js_import_resolution.py passes 61/61 — including the e5a4b51 owned-symbol guard tests (test_alias_import_does_not_remap_an_owned_symbol_id, test_alias_import_preserves_owned_same_line_symbol_edge).
• Regression sweep: failure sets for tests/test_extract.py + tests/test_charmap_encoding.py are name-for-name identical between this branch and clean origin/v8 in my environment (31 pre-existing failures from missing grammar packs on both sides).

Fixes #1983.