bugc: example program-validity test cost grows with instructions times locals #396

The "Program validity" test in packages/bugc/test/examples/examples.test.ts validates every distinct instruction context with Hyperjump. Because bugc lists a function's locals at every instruction, consecutive instructions get mostly distinct, large contexts, so the context cache rarely hits and the cost grows with instructions times locals in scope. advanced/token-registry.bug took 17.4s at O0 on CI (4.7s at O1), which is why #395 raised the per-test timeout from 10s to 30s. The next larger example will cross that bound too.

Options, cheapest first: validate smaller units (each distinct variable entry or pointer inside a context) so they share across instructions; validate only contexts that differ from the previous instruction's; or validate the full context set at one level and the context-free skeleton at the others. Worth timing per example first to confirm the cost is validation and not compilation.