feat(stir): implement STIR protocol as alternative to DEEP-FRI (OPT-S1→S7)

[nol4lej]

Protocol & Core

• Add src/stir.rs: StirProver, StirVerifier, StirRound, StirInnerProof
• Optimize lagrange_eval_at_coset with batch_inverse (Montgomery trick)
• Add phase_time! macro + HYPER_SNARK_PHASE_TIMING instrumentation (5 timing points)
• deep_fri.rs: add FriFoldProof::Stir, runtime protocol dispatch
• config.rs: PROOF_VERSION=18, STIR_QUERY_ROUNDS, active_protocol()

Proof Header (self-describing)

• proof.rs: embed num_queries and pow_bits in Proof struct
• verifier.rs: read pow_bits/num_queries from proof, not env vars

Size Optimizations

• merkle.rs: truncate BLAKE3 digests 24B → 16B (OPT-S1, PROOF_VERSION=12)
• config.rs: compact profile 15q+30PoW (OPT-S2, PROOF_VERSION=13)
• folding.rs / fri.rs / domain.rs: fold-by-8 FRI (OPT-S3, PV=15), reverted in PV=17

Tests

• zk_tests.rs: e2e STIR tests, tampering rejection, STIR < DEEP-FRI size
• Mutex to serialize HYPER_SNARK_PROTOCOL env-var tests
• merkle_tests.rs / unit_merkle_tests.rs: updated for 128-bit digests

Scripts & Docs

• profile-prover.sh: stir / dfri / stir-dfri modes
• docs/timeline/phase-5_proof-size-footprint.md: OPT-S1→S7 results documentd

Fix clippy: doc_lazy_continuation, too_many_arguments, needless_range_loop