[WIP] Trigram based indexer sidecar

[fabxc]

This roughly outlines how an sidecar-based indexer would look like.
First commit is mostly just a big code move to share functionality previously in the store pkg.

It's not fully wired together yet (no API) but one can play with the test skeleton in the index package (BYO log file), which implements detection and indexing of new segments and allows querying them.

The trigram is "only" a pre-filtering optimization, i.e. may yield false positives that need to be checked, which can be anything between 0 and all records of a segment, depending on how well we could extract trigrams from a given regex.

As a result, observed speedups vary from 5-200x. Storage overhead for relatively uniform logs (e.g. webserver logs) appears to be ~70% of the raw data.
Main concern is probably indexing speed at only ~13MB/sec single-threaded.

Probably best to get #120 in before continuing here.