[SizzleUnrlsd]

Summary

The last documented approximation of the analysis, item of #71: a nonlocal write made by a nested function now flows back to the enclosing function.

• Acceptance tests first (test: define nonlocal writes flowing back to the caller), implementation follows.
• Lowering. A store to a name the function declares nonlocal also emits a set_nonlocal effect, kept through SSA like set_global. After a direct call to a nested function it defined, the enclosing function emits one nonlocal_result per name that function declares nonlocal and stores it back into its own variable, so SSA sees the rebinding.
• Summaries. FunctionSummary.nonlocal_writes records, per written name, the parameters and external results the nested function stores into it. The dependence and taint engines evaluate nonlocal_result from the callee's summary and the call's arguments, captured values included, exactly as they evaluate a call's return.
• Cache. The codec carries nonlocal_writes; the format is bumped to 3 so older entries are recomputed.
• The test that documented the limit now asserts the flow; the usage guide states the remaining boundary: a nested function passed around and called elsewhere keeps its writes to itself.

Part of #71.