Fix(Processes): Make subterm hoisting deterministic across processes

[wjrforcyber]

Issue

Shared-subexpression discovery used set.pop(), so unspecified hash-table order—affected by Python's per-process hash randomization—could change which repeated nodes were discovered and hoisted first. This changed synthetic let numbering and serialized e-graph output across otherwise identical processes.

Fix

After integrating current main, _exprs_multiple_parents() uses deterministic ordered DAG traversal with explicit incoming-edge counts, preserving source order and returning each shared node once.

Test

The regression runs the same high-level program in fresh subprocesses with fixed PYTHONHASHSEED values 1, 2, and 3, then asserts that the public saved Egglog transcripts match. The previous set-based traversal produces different transcripts for these seeds.

Changelog

Adds an UNRELEASED entry linking #422.

[codspeed]

Merging this PR will not alter performance

✅ 14 untouched benchmarks

---

_{Comparing wjrforcyber:fix/deterministic-subterm-hoisting (20c7990) with main (8dd2434)}

[saulshanabrook]

Thank you for this!

[saulshanabrook]

@actions-user changelog

[saulshanabrook]

This was fixed in my other branch that integrates a bunch of upstream changes, but keeping this test case, thank you!