Feat: Add BlackBox::Sha256Compression

[shreyas-londhe]

Description

Adds complete SHA256 compression blackbox function support to NoirToR1CSCompiler, enabling efficient SHA256 hashing in R1CS circuits with significant performance improvements through optimized binary operation batching.

Key Innovation: Global Binary Operation Optimization

Implements novel approach to decouple binary operation value computation from constraint generation:

• Collection Phase: add_binop() allocates witnesses, adds WitnessBuilders for immediate value computation, and collects operations
• Constraint Phase: add_binop_constraints() processes all collected AND/XOR operations in a single batch with shared lookup tables
• Benefits: Eliminates redundant digital decompositions and lookup table setups, dramatically reducing memory usage and constraint count

Performance Results

Benchmarked on complete_age_check example:

Metric	Main Branch	SHA256 Blackbox	Improvement
Opcodes	590,297	84,204	86% reduction
Proving Time	22s	~9s	59% faster
Peak Memory	10.4GB	5.25GB	50% reduction
Constraints	3,604,748	2,503,439	31% reduction
Witnesses	5,074,284	3,578,996	29% reduction

Changes Made

Core Implementation

• sha256_compression.rs: New module implementing SHA256 compression algorithm
  • add_sha256_compression(): Main entry point processing 16 input words + 8 hash values → 8 output words
  • 64-round compression loop using SHA256 constants and message schedule

Binary Operation Batching

• binops.rs: Refactored for collection-based processing
  • add_binop(): Collects operations while computing values immediately
  • add_binop_constraints(): Processes all collected operations with shared digital decomposition and lookup tables

WitnessBuilder Extensions

• WitnessBuilder::U32Addition: Computes result = (a + b) % 2^32, carry = (a + b) / 2^32
• WitnessBuilder::And / WitnessBuilder::Xor: Bitwise operations for 32-bit values
• Added corresponding solver logic in prover/src/witness/witness_builder.rs

Integration

• noir_to_r1cs.rs: Added SHA256 compression blackbox function handling with batched binary operation processing

[veljkovranic]

Maybe we can just say here that you need to call add_binop_constrains on every collected_ops Vec that you pass into this method, if we end up needing to add more Binary ops than we did so far

[veljkovranic]

Left a minor comment, otherwise great work!
Really awesome PR :)