sema: run work at the LAUNCH_DMA release path too; log LAUNCH_DMA inventory

The CE pushes UVM's MEMOPS channels submit release through SET_SEMAPHORE_A
plus a LAUNCH_DMA with SEMAPHORE_TYPE=RELEASE_ONE_WORD. That path still
gated the push's collected fills/copies on the semaphore being readable
and behind the payload - the same bootstrap deadlock as the REPORT_SEM
path fixed before: the page-table writes that make the semaphores
translatable were dropped. Run the work unconditionally there as well
(trusted-ring pushes only).

Also log a one-line LAUNCH_DMA inventory (channel, engine, dma word,
modes, offsets, line length, collected op count) for the first pushes
read through rings, which is how the UVM page-table write sequence
(fill 512x8 zero + 16-byte PTE copies + release) was identified.

0 files changed, 0 insertions, 0 deletions