Currently, during vertical compaction, we directly merge the overlapping chunks and the samples can exceed the limit of 120 in a chunk. Chunk needs to be broken down into smaller chunks if it crosses 120 samples.

The piece of code where the fix goes: https://github.com/prometheus/tsdb/blob/d5b3f0704379a9eaca33b711aa0097f001817fc2/chunks/chunks.go#L208-L240