Performance improvement for baseline correction

[surbhipoddar]

Describe the bug Currently, the whole EIC is pulled to detect the baseline. This lowers the performance of the algorithm.

Expected behavior We can limit the number of relevant points around the peak we detect while baseline correction instead of detecting the whole EIC. This will help improve the performance of the algorithm.

[saifulbkhan]

Since we have started using optimization flags on all three platforms, we should see a fair bit of performance increase. There can still be some improvements such as the one suggested above.

[saifulbkhan]

An attempt has been made (available from release v0.8.0) to downsample a signal before calculating its baseline. This should increase browsing and peak detection performance while AsLS Smoothing is turned on.

[shubhra-agrawal]

@chubukov FYI. Since the major critique for AsLSS was around its performance, we have optimized it in v0.8.0. Please let us know if you see an improvement for your datasets.

[saifulbkhan]

Nobody seems to notice any significant slowdown while detecting peaks using AsLS as the baseline algo. We did benchmarks too, it is only a little slower than the default (thresholding as I call it) for smaller number of scans and even outperforms it when the scan density increases.