Write benchmarks so we can improve performance

To be honest, I think what we need are benchmarks rather than tests. There weren't many regressions in the past and writing tests for core functionality will be a lot of work with little benefit. I think writing benchmarks instead and trying to optimize performance and/or memory usage will be much more beneficial

Places to benchmark:

1. Gyro data parsing

2. Gyro integration

3. Calculating frame transform

4. Calculating FOVs for zooming

5. Smoothing algorithm

6. Definitely GPU kernels - we can focus on the SPIR-V implementation, because other ones will probably be replaced by this one soon

There's also the UI (in QML), which can also be profiled and optimized, but not sure how to write automated benchmarks for that

Originally posted by @AdrianEddy in #728

Also, rendering time! There was just a rendering performance regression after ffmpeg update, and it was significant (40% slower!).

I didn't notice it for a few weeks until just now when I compared previous version for some other test.

Having rendering time benchmark will definitely be needed.

The regression was inside ffmpeg, so not related to Gyroflow code, but if we had it automated it would be visible right away

Originally posted by @AdrianEddy in #728