fix(extract): don't report deferred import() as a file cycle

[Synvoya]

What it fixes

A deferred import('./x') was emitted by _dynamic_import_js (graphify/extract.py) as a plain imports_from edge — indistinguishable from a static import. find_import_cycles (graphify/analyze.py) counts every imports_from / re_exports edge as a hard file-cycle edge, so two files that reference each other via one static import + one dynamic import() were reported as a circular dependency (#1241).

Fix

Keep the deferred edge as imports_from — the dependency stays visible in the graph — but tag it "deferred": True, and skip deferred edges in find_import_cycles.

This is deliberately not a retag to dynamic_import: the deferred edge's source is a symbol node (the enclosing function) and its target isn't a materialized file node, so the valid-id prune passes (extract.py:5380 and the sibling passes) only spare it while its relation is imports_from / re_exports. Retagging would drop the edge entirely and lose the dependency; the deferred marker keeps it.

// actions.ts
export async function lazy() { const m = await import("./modal"); return m.openModal(); }
// modal.ts
import { doThing } from "./actions";

• Before: find_import_cycles → [{'cycle': ['actions.ts', 'modal.ts'], 'length': 2, ...}]
• After: [], and the deferred dependency edge is still present (imports_from, deferred=True).

Test

tests/test_js_import_resolution.py::test_ts_dynamic_import_does_not_create_phantom_cycle asserts the deferred edge stays (imports_from + deferred), the real static import is unaffected, and find_import_cycles returns no cycle. Full test_js_import_resolution.py + test_analyze.py pass (99 passed); reverting the fix fails the new test (negative control).

Closes #1241