Add AVX2 double-precision BODFT combine (forward/inverse) and Makefile asm target

[falseywinchnet]

Motivation

• Provide a 256-bit AVX2/FMA double-precision combine implementation for the BODFT path to improve forward and inverse performance and eliminate a final copy in the forward pass.
• Expose assembly emission for the new AVX2 BODFT code via make asm-check and document validation/performance notes.

Description

• Implement AVX2 helpers and vectorized combine routines: added bodft_uzp4_pd, bodft_store4_pd, bodft_rev4_pd, bodft_combine_fwd_avx2_f64, and bodft_combine_inv_avx2_f64 in src/detail/bodft_kernel.hpp and wire them into combine_fwd/combine_inv under BRUUN_LEVEL >= 2 for double (RT size 8).
• Change plan execution to write the final forward pass directly into the caller output buffer and simplify ping-pong buffer switching in both forward and inverse to avoid an extra final copy.
• Add assembly target and variables to Makefile so asm-check emits src/bodft_avx2.s when AVX2 flags are enabled by introducing BODFT_AVX2_ASM and a dedicated rule.
• Add documentation src/detail/bodft_avx2_notes.md describing the AVX2 paths, validation steps, and performance observations.

Testing

• Ran make test BUILD_DIR=build-avx2 CXXWARNFLAGS='-Wall -Wextra -Wpedantic -Werror' which completed successfully.
• Built and executed the BODFT benchmark with make build-avx2/examples/bodft_benchmark BUILD_DIR=build-avx2 && build-avx2/examples/bodft_benchmark to validate correctness and measure timings, which ran successfully.
• Ran make asm-check BUILD_DIR=build-avx2 and inspected build-avx2/src/bodft_avx2.s to confirm AVX/FMA instructions were emitted as expected.

---

Codex Task