feat: improve static analysis — type inference, framework awareness, func refs

[Zarvhos]

Summary

Improves Axon's static analysis pipeline across four dimensions, addressing #55:

• Type inference for receiver resolution: builds per-file {variable: class} mappings from annotations (user: User), parameter types (def f(user: User)), and constructor calls (user = User()). Resolves user.save() → User.save instead of dropping unresolved receiver calls.
• First-class function references: detects handler = my_func and JSX onClick={handleClick} patterns, creating CALLS edges with 0.7 confidence discount.
• Module-level call attribution: calls outside any function body (e.g. app = FastAPI()) are now attributed to the File node instead of being silently dropped.
• Framework-aware entry points & dead code: expanded detection for FastAPI verbs (@app.get/post/put/delete/patch), Next.js data fetching (getServerSideProps, generateStaticParams, etc.), Alembic migrations (upgrade/downgrade), and Pydantic decorators (computed_field, model_serializer, field_serializer).

Also extracts shared path utilities (is_alembic_migration, is_test_file) into a dedicated path_utils module to eliminate duplication.

Benchmark results (4 real-world repos)

Repo	Branch	Files	Syms	Calls	Entry	Dead	FRef
FastAPI Template	main	131	470	649	222	84	—
	feat	131	470	998	145	34	64
Bulletproof React	main	447	493	359	241	81	—
	feat	447	493	879	121	14	282
AI Hedge Fund	main	42	253	264	3	78	—
	feat	42	253	294	2	63	7
Axon	main	137	1630	2754	729	46	—
	feat	137	1698	3228	740	9	131

• Calls edges: +54% to +145% more resolved (type inference + func refs + module-level)
• Entry points: better calibrated (fewer false positives from "no incoming calls" heuristic)
• Dead code: -60% to -83% false positives (framework decorators + Alembic + EXTENDS check)
• No regressions: file counts and symbol counts stable, performance unchanged

Changes by commit

1. feat: add FuncRef, assignment_target, and variable_name to parser models
2. feat: extract func refs, assignment targets, and variable names in Python parser
3. feat: extract JSX calls, func refs, and assignment targets in TypeScript parser
4. feat: add type inference, module-level calls, and func ref resolution
5. feat: add FastAPI, Next.js, and Alembic entry point detection
6. feat: reduce dead code false positives with new exemptions
7. test: add comprehensive tests for static analysis improvements
8. refactor: extract shared path utilities into path_utils module

Test plan

• All 738 tests pass (including ~30 new tests)
• Benchmarked on 4 real-world repos (FastAPI, React, Python ML, Axon itself)
• No regressions in symbol counts, file counts, or performance
• Ruff lint and format clean