Add C benchmark infrastructure to compare current branch vs v0.8.0

[Copilot]

Adds a self-contained C benchmark to quantify solve-time changes on this branch relative to v0.8.0, and applies the micro-optimizations that came out of that analysis.

Benchmark (benchmark/)

• bench_qp.c — standalone C program, no Julia/MATLAB required. Generates dense random QPs at n=10–80, runs 3000 solves per size, takes best-of-7 passes to suppress OS jitter. Links against any libdaqpstat.a.
• README.md — build instructions and two-version comparison workflow.

# Build and compare
gcc -O3 -fassociative-math -fno-signed-zeros -fno-trapping-math \
    -I include -I codegen benchmark/bench_qp.c build/libdaqpstat.a -lm -o bench
./bench

Measured results (identical flags: -O3 -fassociative-math -fno-signed-zeros -fno-trapping-math):

Version	Grand total (5 sizes × 3000 solves, best-of-7)
v0.8.0	~1.083 s
HEAD	~1.108 s (+2.3%)

Small problems (n≤20) are slightly faster; the regression concentrates at n≥60 and is under investigation.

Solver micro-optimisations (src/, include/)

• daqp_dot → static inline (include/factorization.h): was a cross-TU call preventing inlining and vectorisation at all call sites. No LTO required.
• Scalar hoisting (src/auxiliary.c): work->lam_star[i] indexed 4–6 times per outer iteration in daqp_compute_primal_and_fval; hoisted to lam_i.
• LDL_add D-update (src/factorization.c): local scaled variable avoids double-read of work->L[disp]; local sum used directly for singularity check instead of re-reading work->D[n_active].
• Gill-1974 downdate (src/factorization.c): cache work->D[i] in Di to avoid two array reads; reorder w[r]/L[disp] update to break store-load dependency chain.
• daqp_activate_constraints correctness fix (src/auxiliary.c): overdetermined initial working set now correctly distinguishes immutable (equality) constraints that cannot be silently deactivated, rather than blindly clearing all remaining active bits.

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.