Fixed crashes caused by invalid concurrent writes to drv->nextObjId

[thesword53]

I restored the old behavior in nvCreateContext with num_render_targets and I implemented reconfigureDecoderSurfaceCount to reconfigure decoder if the application create surfaces after context creation.
I also rounded up memory size to 4096 in alloc_image as allocation seems to fail in mpv with the 570.133.07 driver:
[drm:__nv_drm_nvkms_gem_obj_init [nvidia_drm]] *ERROR* [nvidia-drm] [GPU ID 0x00002d00] NvKmsKapiMemory 0x00000000e087270f size should be in a multiple of page size to create a gem object

[thesword53]

Hello @elFarto, I modified the pull request to properly address crashes with Firefox 137 when decoding multiple video streams #359. The increment of drv->nextObjId should occur in drv->objectCreationMutex as objects can be created concurrently in multiple threads.

nvidia-vaapi-driver/src/vabackend.c

Line 247 in 14b519d

newObj->id = (++drv->nextObjId);

Firefox 137 seems to use 1 driver context with multiple decode contexts and prior versions are using multiple driver context with 1 decode context. It would explain why we can't observe crashes in Firefox < 137