[SizzleUnrlsd]

Summary

A summary carried the effects of blocks constant propagation proves unreachable: a sink, a return, a value merged at a join, an argument mutation under if False: or while False: were reported at every caller, through any number of project functions.

Change:

• A block proven unreachable has no outgoing state in the summary's dataflow (_DependenceProblem.flow) and is skipped by the final pass. Its stores, calls, returns and writes never happen, and the phis of its successors ignore it.
• Blocks that may run are kept, whatever decides them at run time: a parameter, or a module-level name intraprocedural constant propagation does not resolve.
• Reachability is judged per block. An infeasible edge between two reachable blocks is still followed, which keeps the summary conservative.
• The selection check added in feat(code-injection): report attributes chosen by input and then called #257 is now implied; the invocation check stays.

Verification

• tests/test_summary_reachability.py, new: 11 tests.

  • 6 reproductions fail on main: a sink under if False/while False, a return, a value merged at a join, a mutated argument, and two levels of functions.
  • 5 controls stay reported: a parameter flag, a module-level constant, and the same effects in reachable blocks.

  No existing test or snapshot changed.

• Mutation check: removing either guard fails the new tests.

• Full suite green, apart from the local test_packaging metadata issue. ruff check and mypy clean.

• Corpus: no finding added or removed in any of the 19 regression repositories, nor on mitmproxy (73 findings).

Closes #258