Dense event matching

[bmcfee]

Is your feature request related to a problem? Please describe.

In implementing #1766, i ran into a snag when trying to extend onset backtracking to support multichannel inputs. The problem is as follows:

• multichannel support requires densely encoded (boolean) outputs to indicate detections to avoid ragged output
• backtracking entails matching each detection to the nearest preceding local minimum
• matching is done efficiently using our event matching function. This only supports sparsely encoded events

Describe the solution you'd like

It should be possible to implement an efficient matching algorithm for boolean inputs. Probably the way to do it is to convert the target events to a sparse representation and then use searchsorted, but we shouldn't need to do this explicitly with the source events.

Describe alternatives you've considered

We could hack the onset backtracking code to wrap everything in a dense -sparse-dense conversion, and vectorize accordingly so that it behaves transparently. This might not be as efficient as a direct implementation, but it would work.

[bmcfee]

I gave this a bit more thought, and I'm actually not sure that a direct implementation (ie linear search) would have any benefit over the alternative that would convert dense (booleans) to sparse (indices) and use searchsorted. So I'm leaning toward just doing the latter.

Since this is a pretty niche use case, I think we can just build it directly into the backtracking function and not lift it up to the main API.