Merge #2164

2164: [mtl] Borrowed commands r=grovesNL a=kvark

PR checklist:
- [ ] `make` succeeds (on *nix)
- [x] `make reftests` succeeds
- [x] tested examples with the following backends:

r? @gfx-rs/metallists 

This PR attempts to have lightweight software commands that don't take any heap space or own ObjC objects. In most cases, where a command list is live-recorded and executed once, this should reduce the amount of work we do per command, which is especially important if those commands are thrown away (e.g. because we are not inside a render pass).

My expectation would be to see an improvement in #2161 due to us doing less work. The actual results are somewhat shocking: with v-sync enabled I'm getting the same 59-60 fps as usual. With v-sync OFF, I'm getting between 25 and 50 fps now (which is lower than the previous 50-70). Not sure what's going on, the instrumental profile doesn't give a clue. Please check out the code.

Co-authored-by: Dzmitry Malyshau <[email protected]>

Author: bors[bot]