fix: resolve calls edges through an aliased Python import (#2082)

[Yyunozor]

Summary

from pkg import mod as alias records the file-level imports_from edge correctly, but every downstream alias.func() call is dropped — no calls edge is produced, so the callee looks like dead code even though the import graph looks complete. import pkg.mod as alias regresses the same way. The unaliased form resolves fine.

Reproduction

pkg/gate.py:
    def validate(rows):
        return bool(rows)

pkg/c_alias.py:
    from pkg import gate as m_gate
    def run_plain_alias():
        return m_gate.validate([1])

$ graphify extract . --code-only

c_alias.py --imports_from--> pkg/gate.py is present; run_plain_alias --calls--> validate is missing. Swapping out the alias (from pkg import gate, gate.validate(...)) makes the edge appear.

Root cause

The cross-file member-call resolver's module arm (_resolve_python_member_calls in extract.py, #1883) matches a call receiver against the imported module's own file stem. mod.func() resolves because the receiver equals the stem; alias.func() never does, so the arm bails. The local alias binding is parsed correctly in two places already (_python_imported_names in extractors/resolution.py, and the aliased_import branch of _import_python) but discarded before it reaches the edges the resolver reads.

Fix

Thread the alias through as a local_alias field on the imports/imports_from edge (same transient-hint pattern as target_file, #1814, popped once consumed so it never reaches graph.json):

• _import_python stamps the alias from import pkg.mod as alias.
• _SymbolResolutionFacts.module_imports gains a local_name slot so the from pkg import submod [as alias] submodule path (Package-form imports (from package import module) are not resolved to import edges — creates disconnected test-file islands #1146) carries the binding through.
• The module arm's receiver match now accepts the tracked alias in addition to the module's real stem, keyed per (importing file, target module).

Known limitation

Two different aliases for the same module in one file only resolve the last one registered (one alias slot per importing-file/target-module pair) — not a regression, since the parent resolved neither, and left out of scope here.

Validation

Five regression tests in tests/test_extract.py: the issue's shape (from pkg import gate as m_gate), its try/except-guarded variant, the adjacent import mathlib as m and dotted import pkg.gate as g_alias forms, and the relative from . import gate as r_gate form. All fail on the parent commit with the missing-edge symptom, pass after the fix; one also asserts local_alias never reaches the returned edges.

uv run pytest tests/ -q
3554 passed, 3 skipped, 0 failed

Parent commit, same environment: 3549 passed — delta is exactly these 5 tests. ruff check clean. #2080's calls-edge-direction regression tests pass unchanged.

[Yyunozor]

Landed on v8 as cc70085, with the follow-up coverage in 8c5f585 — closing as superseded. Thanks!