colblk: deduplicate common index block key prefix

We currently store the keys using `RawBytes` in an index block. We tried using `PrefixBytes` but some benchmarks were slower - likely because of the extra complexity of "bundles".

We should add an in-between variant which is like PrefixBytes but with a bundle size of 1, i.e. we just extract the common prefix for the entire block. This should reduce the block size and improve the binary search time when all keys have a large common prefix.

Jira issue: PEBBLE-434