BuildDDG kill computation is quadratic on large switch-heavy functions #18

Describe the bug
BuildDDG (internal/dataflow/reachdefs.go) computes each node's kill set against every definition in the function, which is O(nodes × defs). On very large, switch-heavy generated functions this dominates L3 time: a goyacc-style function with 3,000 case clauses takes 3.9 s, and one with 6,000 takes 15 s, for that single function.

To Reproduce

1. Generate a function with a 6,000-case switch, each case assigning the same few variables.
2. cango -i <dir> -a 3 -o /tmp/out and time it, or profile BuildDDG.

Expected behavior
Kill computation scales with the number of definitions of the same base, not with all definitions in the function.

Logs
Measured by the final review of PR #16 (synthetic probe): 3.9 s at 3,000 cases, 15 s at 6,000.

Additional context

• Fix: index definitions by Path.Base and test kills only within that bucket.
• No real-app run in the evaluation (7 apps incl. Kubernetes) hit this; it is a scaling risk on generated parsers.