VCS assertion VPI: vacuous-success/disabled callbacks don't register; don't key open attempts on cbAssertionStart

VCS W-2024.09, a design compiled with -debug_access+r (concurrent SVA in a module, no -assert compile options). vpi_iterate(vpiAssertion, NULL) returns every assertion, and vpi_register_assertion_cb returns a handle for cbAssertionStart, cbAssertionSuccess, cbAssertionFailure and cbAssertionKill. It returns NULL for cbAssertionVacuousSuccess and cbAssertionDisabledEvaluation.

Consequence: an implication whose antecedent is false fires Start but never fires an end callback. If you keep an 'open attempts' map filled on Start and drained on Success/Failure, it grows without bound. I measured about 857k open entries after 200k cycles with 10 properties: 2,000,029 starts vs 1,142,368 ends.

What worked: - Identify an attempt by (assertion handle, p_vpi_attempt_info->attemptStartTime) on the end callback. - Detect duplicates in O(1): a second start, or a second end, with the same start time in the same tick. - Don't register Start at all unless you need start events. It nearly doubles the callback count. - Check thread ownership with pthread_self(): all callbacks ran on the simulation thread.

Cross-check: Success+Failure callback counts equalled exactly the pass/fail action-block calls of the same properties (1,092,037 + 50,331).

Related: the run-time option '-assert summary' errors (RA-EDNCT) unless the design was compiled with '-assert enable_diag'. If you only want quiet runs for timing, use '-assert quiet' alone.