feat: add value optimization passes (const_prop, algebraic, copy_prop)

[arnoox]

Summary

Adds three value-based optimization passes that analyze and simplify IR value operations:

Pre-lowering passes (on SSA IR with phi nodes):

• const_prop: Constant folding and propagation. Tracks which variables hold known constants and evaluates expressions using Wasm semantics (via herkos-runtime)
• algebraic: Algebraic simplifications. Rewrites x * 1 → x, x & x → x, etc.
• copy_prop: Backward coalescing and forward substitution. Eliminates redundant copies.

Post-lowering passes (on phi-free IR):

• copy_prop (2nd pass): Forwards the assignments that lower_phis inserted into predecessor blocks, reducing further redundancy.

Key improvements:

• Constant folding: Operations with compile-time-known operands are evaluated and replaced with constants
• Dead code exposure: Const prop may eliminate all uses of a variable, exposing dead instructions
• Phi simplification: phi(5, 5) → const(5) is folded before SSA lowering
• Assignment forwarding: Post-lowering copy_prop eliminates intermediate copies created by lower_phis

Implementation notes:

• const_prop uses herkos-runtime functions (wasm_min_f32, i32_div_s, etc.) for spec-compliant evaluation
• IrValue now derives PartialEq to support test assertions
• herkos-core now depends on herkos-runtime for float operation evaluation
• Passes integrate into the existing optimize_ir and optimize_lowered_ir pipeline

Dependencies:

• Depends on PR feat: add wasm float min/max/nearest operations to runtime #28 (runtime float ops) and PR feat: add optimizer infrastructure and post-lowering structural passes #29 (optimizer infrastructure)

🤖 Generated with Claude Code