perf(session-memory): isolate transcript import selection core

[8Dionysus]

Summary

• Extract the read-only Codex transcript discovery/date/size selection path into the standard-library scripts/aoa_session_memory_import.py sibling.
• Keep title, lineage, archive enrichment, and producer callbacks in aoa_session_memory.py; bind the exact sibling source into producer identity.
• Add focused core tests and document the standalone feedback route; update portable validation source lists and evidence-graph wiring.

This PR is intentionally limited to the accepted Session change on top of main (e192414f36bae3b4f9b2b0a0adadfd92a473837c). It does not include the cold-root locator change, private session data, generated runtime state, installation, or deployment changes.

Validation

• tests/test_session_memory_import_core.py plus validation graph/scheduler tests: 33 passed.
• Loader/source-identity/static selector: 4 passed.
• python3 -m py_compile scripts/aoa_session_memory.py scripts/aoa_session_memory_import.py: passed.
• python3 scripts/generate_decision_indexes.py --check: passed (101 records, 6 indexes current).
• git diff --check: passed.
• python3 scripts/audit_public_tree.py --root . --fail-on blocking: passed (blocking=0; 24 known review findings only).
• A controlled polarity mutation of the size-lane dispatch failed the focused contract test as expected, demonstrating the new test boundary.

The full portable suite was not repeated locally after the bounded focused checks; this PR relies on the repository CI full-suite gate for that separate claim. No speedup is claimed from the paired whole-command sample (it showed no material difference); the measured benefit is an independently testable import core without compiling the monolith.