Commands batches and timing

[benjamin-heasly]

This branch represents a bunch (!) of recent work, including:

• refactoring to an OOP command model in the Metal App
• supporting queued batches of commands
• gathering detailed frame timing for each command, including GPU timestamps

For an overview of the design changes, please see the new Metal App readme.

To run some existing tests, see the new timestamps, and compare timing of normal vs batched commands, you can try these:

mglTestRenderPipeline('runTests=1', 'batchMode=0')
mglTestRenderPipeline('runTests=1', 'batchMode=1')
mglTestRenderPipeline('runTests=2', 'batchMode=0')
mglTestRenderPipeline('runTests=2', 'batchMode=1')
mglTestRenderPipeline('runTests=3', 'batchMode=0')
mglTestRenderPipeline('runTests=3', 'batchMode=1')
mglTestRenderPipeline('runTests=4', 'batchMode=0')
mglTestRenderPipeline('runTests=4', 'batchMode=1')
mglTestRenderPipeline('runTests=5', 'batchMode=0')
mglTestRenderPipeline('runTests=5', 'batchMode=1')

[benjamin-heasly]

Today I also added a command to sample GPU and CPU times together, and Matlab code to use these samples to convert GPU timestamps to CPU time.

The tests with mglTestRenderPipeline should still work the same as above.