[processor/deltatocumulativeprocessor]: fix incorrect bucket counts when downscaling exp histograms

[edma2]

Description:

The downscaling algorithm collapses bucket by pairing neighbors into new buckets. The simple case occurs when there is an even number of buckets and offset is zero or even.

1 1 1 1
 2   2

When there is an odd number of buckets, the last bucket is accounted for by incrementing the size variable to ensure an extra loop iteration.

1 1 1
 2   1

When offset is odd, the first bucket is accounted for by special handling in the loop.

ø 1 1 1
 1   2

However, when there is an even number of buckets, AND an odd offset, this happens:

ø 1 1
 1   0

The collapsed buckets are incorrect because the buckets sum up to 1 (1+0) instead of 2 (1+1).

Handle this edge case by incrementing the size variable when offset is odd

ø 1 1
 1   1

Link to tracking Issue:

Testing:

Added some test cases to showcase the bug.

Documentation:

[linux-foundation-easycla]

The committers listed above are authorized under a signed CLA.

• ✅ login: edma2 / name: Eugene Ma (f447f0a, fbf55d2, f02c5dd, 045ce5c, 964a174, 361499b)

[jesusvazquez]

LGTM

[sh0rez]

@edma2 please add a changelog entry: https://github.com/open-telemetry/opentelemetry-collector/blob/main/CONTRIBUTING.md#adding-a-changelog-entry

otherwise LGTM

[edma2]

@sh0rez thanks, added a changelog entry

[edma2]

@sh0rez @jpkrohling thanks for your reviews and approvals, can we merge it now?