Fix memcpyopt to allow multiple passes of optimisations

[xunilrj]

Description

This PR is another prelude to #7646.

We discovered that memcpyopt does not work when we run it multiple times. The cause is that we bail the optimisation when a symbol has more than one write. This is correct, but we were only considering 'store'.

Now we are correctly also considering all other ways to write into "symbols".

Checklist

• I have linked to any relevant issues.
• I have commented my code, particularly in hard-to-understand areas.
• I have updated the documentation where relevant (API docs, the reference, and the Sway book).
  • If my change requires substantial documentation changes, I have requested support from the DevRel team
• I have added tests that prove my fix is effective or that my feature works.
• I have added (or requested a maintainer to add) the necessary Breaking* or New Feature labels where relevant.
• I have done my best to ensure that my PR adheres to the Fuel Labs Code Review Standards.
• I have requested a review from the relevant team or maintainers.

[xunilrj]

@cursor BugBot run

[cursor]

Here's my BugBot analysis of PR #7657:

CI Status: All green. 34 checks passed, 0 failed. The only pending item is policy-bot: master (a review policy check, not a code issue). 5 checks were skipped (publish-related, only triggered on merge).

Local Build & Test Results:

• cargo build --workspace — compiles successfully
• cargo test -p sway-ir — all 37 tests pass (18 unit + 19 integration including memcpyopt and memcpy_prop)
• cargo clippy -p sway-ir — only 1 pre-existing minor warning (unrelated to this PR)
• cargo test -p sway-core --lib — all 20 tests pass

Code Review Summary:

The changes are sound. The two key modifications are:

1. memcpyopt.rs — The stores_map now correctly tracks all write instructions (Store, MemCopyVal, MemClearVal, MemCopyBytes), not just Store. This fixes the bug where multiple passes of the optimization would fail because the "single write" check didn't account for MemCopy/MemClear operations that are themselves produced by prior optimization passes.

2. pass_manager.rs — The pass group now runs in a loop (up to 2 iterations), re-running all passes if any pass made modifications. This enables cascading optimizations that only become available after a first pass transforms the IR. The early exit when nothing is modified ensures no performance regression for code that converges in one pass.

No bugs or issues found. The PR is ready for human review/approval.

  

[ironcev]

👍

[ironcev]

Great to see those gas reductions in tests :-) Two nitpicks in comments for retaining initial clarity, otherwise 👍 👍