fix: Fix panic on requesting out-of-order Pattern samples

[benclive]

What this PR does / why we need it:
This PR fixes a panic that can occur when samples are out-of-order in the Pattern chunks.
This happens because the calculated slice length can be negative if the timestamp at hi-1 is less than the timestamp at lo due to subtracting currentStep (generated from timestamp@lo) from the timestamp@hi-1.

	currentStep := truncateTimestamp(c.Samples[lo].Timestamp, step)
	aggregatedSamples := make([]logproto.PatternSample, 0, ((c.Samples[hi-1].Timestamp-currentStep)/step)+1)

Implementation Notes

• I opted to return nil if the timestamps are out-of-order which may be incorrect behaviour (but does solve the panic).
• The samples should never be out of order now since I added a sort to the Add method.
• An alternative is to simply reject samples that are older than the latest during Add - let me know if you prefer this approach.

[ashwanthgoli]

this might not cover all the cases since the in-between samples could be out-of-order?

[benclive]

Good point - I'll add some more test cases

[benclive]

Rewrote this so it will reject out-of-order samples for the Pattern instead of trying to sort this chunk.
Firstly, this makes the sample generation code much simpler. Secondly, this is a fairly rare case and since Patterns don't need to be 100% exact, its OK to simply drop these values.

[ashwanthgoli]

lgtm