Fix(Processes): Make subterm hoisting deterministic across processes

[wjrforcyber]

Issue

serialize() output (and the internal $__expr_N let-numbering / e-class ids) differ between processes for byte-identical input programs.

_exprs_multiple_parents() traversed the expression graph with a set of TypedExprDecl objects and .pop()ed from it. Popping from a set of objects follows id()-based hashing — i.e. memory addresses — so the traversal order (and with it the order/decision in which shared subterms were hoisted into $__expr_N let-bindings) varied per process. Identical programs were therefore lowered into different (equivalent) e-graphs.

Fix

Traverse with a deterministic LIFO worklist (list + .pop(), children via .extend()) instead of a set.

Test

Builds an expression with many shared (multi-parent) subterms, lowers it in 3 fresh subprocesses, and asserts the serialized output hash is identical.

[codspeed-hq]

Merging this PR will not alter performance

✅ 12 untouched benchmarks
⏩ 2 skipped benchmarks¹

---

_{Comparing wjrforcyber:fix/deterministic-subterm-hoisting (c33c4f1) with main (6b2016e)}

Footnotes

1. 2 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩