perf: chunk and process playback segments concurrently

[boenshao]

The /list API can be a bit slow when there are lots of recordings that need to be processed.

The concatenation procedure can be split up and executed concurrently (and even parallelly).

I hard-coded the splits into 4, as most modern platforms have at least 4 cores.

But it's possible to introduce a more sophisticated way to determine the preferred number, either by the number of files or the total size of the files. We can gather the statistics in os.Walk.

Yet I believe the overhead of goroutine is negligible in most cases, so a static strategy is probably alright.

With this patch, on my AlderLake laptop, the response time is about 3x faster when parsing 2X GBs of recordings.

[aler9]

Hello, the main bottleneck of the /list endpoint was caused by the fact that the duration of each segment was computed by reading every part of the segment and summing their duration. This is fixed in #4096.

The idea behind this PR is that performing parallel reads on a disk improves performance. I've gathered some documentation on the topic, in particular this article is quite interesting:
https://pkolaczk.github.io/disk-parallelism/

Basically, parallel readings improve performance with SSDs but decrease performance with HDDs when readings are sequential. And HDDs are still popular for 24/7 video recordings.

Therefore: this patch is interesting but i'd like to take it into consideration after evaluating whether #4096 is enough for providing decent performance in most cases, which will take some time (it applies only to new segments and not to existing ones).

[aler9]

This PR has an additional issue: it prevents more than 4 segments from being grouped together. The /list endpoint is supposed to provide timespans as long as possible, in order for them to be represented in a UI, not the segment list, which can be quite large.

This is replaced by #4102 which parses segments in parallel but performs concatenation in series.

[github-actions]

This issue is mentioned in release v1.11.0 🚀
Check out the entire changelog by clicking here