Python get_method_bodies lacks the None-code omission filter (twin of the TS #298 rule)

[rahlk]

PROBLEM

The TS bulk accessors (#298) ship get_method_bodies with a mechanical omission rule: entries whose code is None are omitted, so every returned value is a real str, documented on both the ABC and the facade. Python's implementations pass c.code straight through with no filter, and the docstrings make no such promise — a Dict[str, str] contract violation whenever a PyCallable carries no source text. Found during #298's final review.

SCOPE BOUNDARY

get_method_bodies on the Python side only — the in-memory backend and the Neo4j backend, plus the two docstring layers. Not the other bulk accessors, and not the TS side, which already has the rule.

GOALS

• In-memory (cldk/analysis/python/codeanalyzer/codeanalyzer.py): filter on c.code is not None.
• Neo4j (cldk/analysis/python/neo4j/neo4j_backend.py): add AND c.code IS NOT NULL to the MATCH, index the row directly instead of r.get("code").
• Document the rule on the Python ABC and the facade, as TS does.
• Land on both release lines.

CAVEATS AND KNOWN RISKS

• Latent, not reproduced. It is unconfirmed whether codeanalyzer-python ever actually emits code=None; the fix closes a declared-contract hole either way, but no failing case has been observed.
• The two backends must agree — fixing only the in-memory path recreates the dual-backend divergence this rule exists to prevent.

DEFINITION OF DONE

• Every value returned by get_method_bodies is a real str on both backends, with a callable carrying code=None omitted rather than yielding None.
• Python and TS return the same shape for the same input.
• Rule stated in the ABC and facade docstrings.

STATUS

Fixed on release/2.0 by #305 (merged). Still open for main — the "both release lines" goal is unmet, and GitHub did not auto-close because #305 did not target the default branch.