Rework GPU Timing Stats

[bjornbytes]

• Add a new async lovr.graphics.getTiming function that returns a table of timestamps for the last GPU submission, keyed by pass.
• The timestamps are formatted like this:

{
  [pass1] = {
    duration = t,
    start = t,
    finish = t,
    submit = t
  },
  [pass2] = {
    duration = t,
    start = t,
    finish = t,
    submit = t
  }
}

• The start/finish times are in the same units/epoch as lovr.timer.getTime, so you can correlate it with other CPU times.
• Not every GPU can correlate timestamps, so start/finish may be nil. But nearly every desktop GPU supports it.
• Deprecate gpuTime/submitTime in Pass:getStats.
• Main goals are:
  • Make it easier to get stats for multiple passes
  • Use an async function to make it easier to correlate stats to a particular submit instead of updating the pass stats asynchronously
  • Use the start/finish times to see when a Pass actually ran (measure CPU-GPU latency, see if passes are overlapping on GPU)
  • Use 64 bit queries internally, because that's all WebGPU supports

Example:

lovr.graphics.setTimingEnabled(true)

function lovr.draw(pass)
  --

  lovr.graphics.submit(pass)

  lovr.task.start(function()
    for pass, t in pairs(lovr.graphics.getTiming()) do
      print(pass, t.duration, t.start, t.finish, t.submit)
    end
  end)

  return true
end