F6: new surface — entrypoints, L3/L4 accessors, AnalysisLevel L1–L4

[rahlk]

PROBLEM

Schema v2 carries facts the facade cannot expose today: framework entrypoints and the L3/L4 program graphs. This issue adds the ADDITIVE new surface — nothing existing moves.

SCOPE BOUNDARY

Additive only. No existing accessor changes (that is F3's frozen surface). Query-core verbs are out of scope (#155, ships 2.1.0) — these are plain accessors over the new facts.

GOALS

1. Implement get_entry_point_methods / get_service_entry_point_methods (today: NotImplementedError) from v2 framework / detection_source / route_path / http_methods and the Entrypoint marker (absorbs feat(typescript): implement entry-point detection (currently raises NotImplementedError) #199).
2. L3/L4 accessors: expose cfg / cdg / ddg / summary (per-callable) and param_in / param_out (application scope), on BOTH backends.
3. AnalysisLevel grows L1-L4 (documented shift 7); the level->flag mapping is re-derived from max_level, never sniffed.

CAVEATS AND KNOWN RISKS

• Levels 3/4 are opt-in and expensive; degrade honestly when max_level < requested (this is the substrate the 2.1.0 query core's capability model builds on).
• Entrypoint coverage depends on each analyzer's level-2 finders; assert on fixtures per framework actually supported (express/nest for TS; py frameworks as codeanalyzer-python lands them).

DEFINITION OF DONE

• Entrypoint fixtures pass per supported framework.
• L3/L4 accessors validated against sample graphs on both backends; requesting a level above max_level produces the documented honest-degrade behavior, not an exception.
• feat(typescript): implement entry-point detection (currently raises NotImplementedError) #199 closed.

Part of #238