Forward repeated finalize storage reads

[Kuhai9801]

Motivation

Fixes part of #29343.

Lowered finalize code can contain repeated identical static storage reads, such as repeated Mapping::get_or_use, Mapping::contains, or vector length reads. After CSE, these reads are still separate lowered storage intrinsics, so DCE cannot remove the duplicated read work.

This PR adds a separate StorageReadForwarding pass after CSE and before DCE. It reuses repeated local static storage reads when the mapping/key/default operands are stable atoms.

The pass is conservative:

• forwards _mapping_get, _mapping_get_or_use, and _mapping_contains
• invalidates tracked reads across mapping writes/removes, final runs, calls, dynamic ops, assert barriers, and unguarded branch exits
• exposes branch-local aliases only to matching arms of same-condition SSA join ternaries immediately after the branch
• keeps dynamic storage reads out of scope because they include target program/network operands and need a separate invalidation model

Test Plan

Added regression coverage:

• storage_read_forwarding/repeated_mapping_reads
• storage_read_forwarding/repeated_vector_len
• storage_read_forwarding/assert_barrier
• storage_read_forwarding/conditional_branch_reads
• compiler/storage/external_same_named_mapping_reads

The conditional-branch coverage includes the same-condition SSA join case and negative cases for different conditions and opposite-branch operands.

Verification performed:

• git diff --check
• conflict-marker scan over changed files
• merge simulation against origin/master
• CircleCI PR checks

Related PRs

Part of the #29343 optimizer cleanup series.

Related PRs:

• Forward aliased aggregate fields in SSA const propagation #29543
• Absorb redundant atom-only SSA ternaries #29544, which closes Collapse redundant Optional wrapping around storage vector access (option_lowering + ssa_const_propagation) #29343