config: change default segment size to 128MiB

[jcsp]

This size is a more suitable default, to have
finer granularity of disk usage, and finer
granularity of tiered storage read cache when
tiered storage is in use.

For systems that write high bandwidths to a single partition, this will increase the number of segment rolls, but should not outright break anything unless the system was already very close to limits of e.g. file handles or memory.

Fixes #8170

Backports Required

• none - not a bug fix
• none - issue does not exist in previous branches
• none - papercut/not impactful enough to backport
• v22.3.x
• v22.2.x
• v22.1.x

UX Changes

None

Release Notes

Improvements

• The default log segment size is now 128MiB, decreased from the previous 1GiB.

[jcsp]

I considered updating compacted_log_segment_size as well, but that feels risky:

• If we scaled it proportionally to log_segment size, the segments would be 32MiB, which is very small.
• Any decrease in the segment size for compacted topics will decrease the efficiency of self-compaction within a segment (as we would have a smaller chance of repeated keys showing up in the same segment). Self compaction is important for tiered storage topics, as this is the main chance for reduction before data is re-uploaded post compaction (we may not upload subsequent compaction passes).

[jcsp]

/ci-repeat 5

[jcsp]

Test results from repeated run:

• Two instances of CI Failure (assert in persisted_stm) in EndToEndTopicRecovery.test_restore #8293
• One instance of CI Failure (decommissioning stopped making progress) in NodesDecommissioningTest.test_flipping_decommission_recommission #8301
• One failure of EndToEndShadowIndexingTestWithDisruptions that cannot result from this PR, because that test sets an override for segment size.