Incorporate codeanalyzer-python v1.6.0 #427

A new codeanalyzer-python release (v1.6.0) is published.
Update the codeanalyzer-python pin in this repo and adapt the
integration to the changes below (see PyCodeanalyzer._run_analyzer).

[1.6.0] - 2026-10-11

Changed

• Level-4 analysis uses about 7x less memory, with the same output (#224). On a
  6,685-module project (Odoo 20.0), peak RSS fell from 42.6 GB to 5.7 GB for JSON output,
  and from 65.1 GB to 5.5 GB for --emit neo4j. On the same input, analysis.json and
  the analysis cache have the same content as 1.5.4. The Neo4j snapshot has the same
  statements, but edge batches can come in a different order. Representation and lifetime
  changes only; no analysis logic changed:
  • The schema's leaf records are slotted stdlib dataclasses, not pydantic models. This
    covers Span, BodyNode, the cfg/cdg/ddg/summary/param_* edges, PyCallEdge,
    PyCallsite, PyCallArgument, PySymbol, PyComment, PyVariableDeclaration,
    PyCallableParameter, PyDecorator and PyImport. The containers (PyModule,
    PyClass, PyCallable, PyApplication, Analysis) stay pydantic models.
  • analysis.json and the cache are written one module at a time, not as whole-tree
    copies. The edge lists are written in chunks. The JSON now uses compact separators
    (only whitespace differs).
  • ASTs, PDG data and the level-4 IR are released as soon as nothing needs them.
  • The Neo4j graph.cypher snapshot is written in two passes. Nodes are written first,
    sorted as before. Edge batches are written as they fill.
  • Ray runs as a private local cluster with object spilling turned off. It is shut down
    after the symbol-table phase. CANPY_RAY_OBJECT_STORE (bytes, default 1 GB) caps the
    object store, and CANPY_RAY_CPUS caps the workers.
• In-process API: the leaf types above no longer have .model_dump() or
  .model_validate(). Use dataclasses.asdict(...) or the constructor. The
  codeanalyzer.schema helpers (model_dump, model_copy, ...) and the container
  models handle them transparently.

Removed

• BREAKING: Python 3.9 and 3.10 are no longer supported; the analyzer needs Python 3.11+
  (requires-python = ">=3.11"). The level-4 memory cut (#224) makes the schema's
  per-edge and per-statement records slotted stdlib dataclasses. Only pydantic v2 validates
  and serializes those the same way as models, and below 3.11 the dependency pins resolved
  to pydantic v1 (with ray==2.0.0). The python_version < '3.11' pin set, the tomli
  backport, the pydantic-v1 branches of the codeanalyzer.schema helpers, and the 3.10
  compat release job are gone. codeanalyzer.schema.PYDANTIC_V2 stays importable and is
  always True. Migration: install the analyzer under Python 3.11 or newer.