runtime: support bpf_ktime_get_boot_ns

[yunwei37]

Summary

• implement and register helper ID 125 (bpf_ktime_get_boot_ns)
• use CLOCK_BOOTTIME, with CLOCK_MONOTONIC as a portability fallback
• exercise the helper through the LLVM JIT helper regression program

Fixes #608

Validation

• cmake -B build -DCMAKE_BUILD_TYPE=Release -DBPFTIME_ENABLE_LTO=NO -DBPFTIME_LLVM_JIT=YES -DBPFTIME_ENABLE_UNIT_TESTING=1
• cmake --build build --target bpftime_runtime_tests -j$(nproc)
• ./build/runtime/unit-test/bpftime_runtime_tests 'Test helpers' (1 test case, 5 assertions passed)
• full runtime test binary: helper regression passed; 75/77 test cases passed, with the two remaining failures requiring host filesystem/kernel BPF privileges (Test shm progs attach, Test tail calling from userspace to kernel)

[yunwei37]

Exact-head review is complete for 8c6e281.

The first pass identified a runtime edge case where CLOCK_BOOTTIME could be available at compile time but rejected by the running kernel. Commit 8c6e281 resolves it by zero-initializing the result and falling back to CLOCK_MONOTONIC on failure. Two final independent reviews found no blocking issues: helper ID/signature/registration are correct, the LLVM JIT regression covers the original unresolved-helper path, and the two-file 20-line scope remains minimal.

The focused test was rebuilt and rerun on the final head: 1 test case and 5 assertions passed. CI has now reached a terminal state with 141 successful checks, 2 expected skips, and no failed checks. The two GPU jobs were cancelled after waiting for unavailable self-hosted runners without executing any repository steps. The PR is mergeable and has no unresolved review threads; the remaining decisions are GPU-runner availability and maintainer review.

AI-generated response; a maintainer will review and follow up later