Add inverse-only benchmark column and enable scheduled inverse for wide x86 (BRUUN_LEVEL <= 3)

[falseywinchnet]

Motivation

• Expose inverse-only performance measurements separately from round-trip timing to make inverse optimizations visible in benchmarks.
• Enable the scheduled inverse traversal for wider x86 ISAs (AVX2/AVX-512) after fused d4 inverse improvements made the flat schedule competitive.

Description

• Added an inverse_ns field to Result and added an inverse-only timing measurement that runs plan.inverse after priming with plan.forward in examples/benchmark.cpp.
• Printed the new Inv_ns column in the benchmark header and output by extending the format and calling print_ns(result.inverse_ns).
• Switched the inverse scheduling gate in src/detail/bruun_kernel.hpp from #if BRUUN_LEVEL == 1 to #if BRUUN_LEVEL <= 3 to enable the scheduled inverse for BRUUN_LEVEL 2 and 3 builds.
• Updated src/detail/inverse_acceleration_notes.txt to record the decision rationale and to note that the benchmark now prints the Inv_ns column.

Testing

• Built and ran the examples/benchmark program and verified the new Inv_ns column appears and reports timings as expected.
• Ran the project's automated test suite with ctest and the tests completed successfully.

---

Codex Task