v2 Stage 1: schema foundation + L1

[rahlk]

Part of #69.

Stage 1 — v2 schema foundation + L1

Introduce the v2 model layer, the can:// id builders, source-per-module + byte-offset spans, the L1 body (call nodes), the JSON envelope, and the Neo4j L1 re-key. After this stage, -a 1 emits a v2-shaped tree in both projections. max_level=1.

Branch: feat/v2-stage1-schema-l1

Tasks (bite-sized TDD, detailed in the plan):

1. Byte-offset Span model + line/col→byte helper (schema/py_schema.py).
2. can:// durable-id builders (schema/ids.py) — locks the name(argnames) param-encoding decision.
3. v2 model layer — BodyNode, split edge models (Cfg/Cdg/Ddg/Summary/ParamEdge), Analysis envelope; reshape PyApplication/PyModule/PyClass/PyCallable; dissolve PyProgramGraphs/PyFunctionGraphs/PySDGEndpoint/PySDGEdge/PyGraphNode/PyParamNode.
4. Capture source per module + byte-offset spans in the symbol-table builder; drop the per-node code field.
5. Stamp can:// ids across the tree (schema/assign_ids.py, returns signature→can://id).
6. L1 body call nodes from call_sites (callee null; schema/l1_body.py).
7. Emit the v2 Analysis envelope from the CLI; analyze() returns Analysis.
8. Canonical-schema conformance gate helper + L1 assertion (test/conftest_v2.py).
9. Neo4j L1 re-key onto can:// ids (two-projection agreement).
10. Stale-v1 cache detection (rebuild instead of crash).

Gate (L1): -a 1 output round-trips the v2 models and passes assert_conformant(..., 1); symbol_table keys relative; get_method_body slices source via span bytes; cache reuse works and a stale v1 cache rebuilds; JSON callable ids == Neo4j merge keys.