core/bloombits: handle non 8-bit boundary section matches

[karalabe]

In our bloombits matcher, we maintain matched blocks via a []byte bitset. After filtering the blocks and generating the final bitset, we iterate over the block interval and check whether they match according to the bitset.

Since the bitset contains bytes, if a bitset element is zero, we can skip ahead 8 blocks, since none of them will match. This has a subtlety that we got wrong. If we start filtering from a block number that's not a multiple of 8, then we can't skip ahead 8 blocks when finding a zero bitset element, only 8 - number % 8 elements, since skipping 8 would overlap with the next byte in the bitset.

This PR fixes it by only skipping 8 blocks if we're at a bitset byte boundary, otherwise just increment blocks 1-by-1 until we do reach a byte boundary.

Fixes #15309.

[holiman]

LGTM

[slavik0329]

Thank god for this fix. Should be released. It's a pretty big issue

[karalabe]

Will be on Monday.