Timing assertion state is header-static, non-thread-safe, and not cross-translation-unit

[xross]

• timing state (timing_blocks, head, tail) is static in header, creating one instance per translation unit
• xassert_timing_start() in one source file cannot match xassert_timing_end() in another
• duplicated state memory across files
• mutable shared state is unsynchronized, so concurrent/threaded or multi-core usage can produce incorrect results
• expected fix direction:
  • Move timing state out of the public header into a single implementation unit (.c/.xc) so there is one shared instance.
  • Guard all timing state mutations/lookups (timing_blocks, head, tail) with the same lock.