…ze analysis

The counter-based approach (complexity_score += 1) couldn't replicate
the original mccabe library's graph formula (E - N + 2). This caused
188 functions to have different scores on astroid alone, almost all
under-counted because the counter misses how paths merge at try/except
blocks and other constructs.

Replace with a proper control flow graph that tracks edge and node
counts, then computes E - N + 2. Only 2 expected differences remain
vs the original mccabe library: match/case (unsupported in mccabe
v0.7.0) and template strings (Python 3.14), both improvements.

Key design: no PathNode objects needed — just integer counters for E
and N. Simple statements are skipped since collapsing linear chains
doesn't change E - N + 2. All performance optimizations retained
(dict dispatch, skip expressions, walk astroid directly).

The dispatch dict mapping node types to visitor methods is now created
once at module level using unbound methods, instead of being rebuilt
for each PathGraphingAstVisitor instance. Callers pass self explicitly.

Eliminates ~18K dict allocations (12 entries each) per ansible run.

Speeds up attribute access for _num_nodes, _num_edges, _tail, _active
and graphs by using C-level struct offsets instead of __dict__ lookups.
These attributes are read/written millions of times during traversal.

Eliminates ~187K method calls per ansible run by inlining the subgraph
parse logic directly into the two callers. For if/for/while (no except
handlers), the extra_blocks loop is dead code and removed entirely.

Astroid nodes are hashable by identity, so we can use them directly as
dict keys. This eliminates ~107K id() builtin calls per ansible run
and simplifies the graphs dict from {int: (result, node)} to
{node: result}.

merge _visit_classdef/_visit_with, use _walk_body from checker

- Store complexity as plain int instead of _ComplexityResult object,
  eliminating ~107K object allocations and method calls.
- Cache self._walk_body as local variable in hot methods to avoid
  repeated bound-method creation via descriptor protocol.
- Merge _visit_classdef and _visit_with into _visit_body since both
  just recurse into node.body.
- Use visitor._walk_body(module.body) from the checker instead of
  a dispatch loop, saving ~18K dispatch calls per run.

…elling

- Add top-level `is_toplevel` handling to `_visit_match` (was an early
  return, inconsistent with `_visit_subgraph` and `_visit_try`)
- Add functional test cases for module-level try/except and match
- Fix spelling issues in comments and docstrings flagged by spell checker

…lling check