fix(obserivation): fix missed hitmap updates, fix drop order

[MrCroxx]

I hereby agree to the terms of the RisingWave Labs, Inc. Contributor License Agreement.

What's changed and what's your intention?

FIx:

• Take block seek read into hitmap's account.
• Make sure hitmap is dropped before the block holder.

Checklist

• I have written necessary rustdoc comments
• I have added necessary unit tests and integration tests
• I have added test labels as necessary. See details.
• I have added fuzzing tests or opened an issue to track them. (Optional, recommended for new SQL features Sqlsmith: Sql feature generation #7934).
• My PR contains breaking changes. (If it deprecates some features, please create a tracking issue to remove them in the future).
• All checks passed in ./risedev check (or alias, ./risedev c)
• My PR changes performance-critical code. (Please run macro/micro-benchmarks and show the results.)

• My PR contains critical fixes that are necessary to be merged into the latest release. (Please check out the details)

Documentation

• My PR needs documentation updates. (Please use the Release note section below to summarize the impact on users)

Release note

If this PR includes changes that directly affect users or other significant modifications relevant to the community, kindly draft a release note to provide a concise summary of these changes. Please prioritize highlighting the impact these changes will have on users.

[wenym1]

Why not just update the local hitmap instead of creating a new one?

[MrCroxx]

The block binary search requires the iterator's immutable reference, and the hitmap cannot be updated in the closure for it requires the mutable reference. (I forgot to restore this diff line, let me fix it.)

[wenym1]

Looks like the immutable reference is required to call decode_prefix_at of self. However, looking into the implementation of decode_prefix_at, it only uses self.block. Therefore, if we change to implement decode_prefix_at as method of BlockHolder, we don't need the immutable reference of self here. And it makes sense to me to make decode_prefix_at a method of a single block instead of a method of BlockIterator.

[wenym1]

Or we can just inline the call of decode_prefix_at as

KeyPrefix::decode(
    &mut &self.block.data()[offset..],
    offset,
    key_len_type,
    value_len_type,
)

[wenym1]

Looks like the offset information is duplicated here, once in slice range, and once as method parameter.

Maybe we can change the first parameter of decode from buf: &mut impl Buf to &[u8] and we generate the &mut &self.block.data()[offset..] inside the code of KeyPrefix::decode so that we remove the duplicated offset information in method call.

[MrCroxx]

+1 for inline call, but I think duplicated offset sounds okay here, it will unify the encode/decode parameters of KeyPrefix.

[wenym1]

The current way of report hitmap leverages the drop of local hitmap, but the correctness of the drop of local hitmap should depend on the correct implementation of the drop of BlockIterator. If so, I think we'd better decouple hitmap and local hitmap, which means we won't associate a hitmap the local hitmap, and the local hitmap won't implement its drop. We only manually apply the local hitmap to the block hitmap in the drop of BlockIterator.