perf: reuse transaction cursor and batch fetch block indices in Transactions segment (no mapping)

[aleexeyy]

This PR addresses the same issue as #19079, but corrects a misunderstanding from my previous PR.

In my first PR (#19196) I explicitly mapped BlockBodyIndices to BlockNumbers, thinking it was necessary to preserve the mapping. After further investigation, it turns out this is unnecessary, because the block_body_indices_range method already returns indices sequentially and stops at the first missing block. Therefore, we can implement the optimization without introducing a new API method.

Note: I have a concern about using usize to enumerate a u64..u64 range. On 32-bit systems, if the range is very large, it could potentially overflow. I need to get your opinion on that, maybe I am wrong.

Changes in this PR

• Reuse the transaction cursor across all blocks in the Transactions segment.
• Batch fetch block body indices for all blocks in the range.
• Consume transactions only up to the last transaction number of the current block.

Expected Benefits

• Reduced overhead during static file writes, especially for large block ranges.
• Simpler implementation without an extra mapping step.

References

• Original issue: closes [[#19079](https://github.com/perf: reuse transaction cursor and batch fetch block indices in Transactions segment #19079)](perf: reuse transaction cursor and batch fetch block indices in Transactions segment #19079)
• First PR with mapping: [[#19196](https://github.com/perf: reuse transaction cursor and batch fetch block indices in Transactions segment #19196)](perf: reuse transaction cursor and batch fetch block indices in Transactions segment #19196)

[shekhirin]

We decided to remove the transactions segment altogether, as it's unused now #19209