venus: add logging and raise VKR_CS_DECODER_TEMP_POOL_MAX_SIZE to 1GB

VKR_CS_DECODER_TEMP_POOL_MAX_SIZE is to avoid integer overflows and to catch bogus allocations (e.g., the guest driver encodes an uninitialized value). In practice, the largest allocations we've seen are from vkGetPipelineCacheData and are dozens of MBs.

This fixes Rise of the Tomb Raider whose vkGetPipelineCacheData calls can exceed 32MB.

/cc @zzyiwei @ryanneph