re_datastore: index bucket compaction

[teh-cmc]

(mentioned by @jondo2010 during retro)

Compact sparse index buckets (because they were filled on a per-component basis rather than multiple components at a time).

Basically automatically going from this

┌──────┬───────────┬───────┬───────────┐
│ time ┆ instances ┆ rects ┆ positions │
│ ---  ┆ ---       ┆ ---   ┆ ---       │
│ str  ┆ u64       ┆ u64   ┆ u64       │
╞══════╪═══════════╪═══════╪═══════════╡
│ #42  ┆ null      ┆ 1     ┆ null      │
├╌╌╌╌╌╌┼╌╌╌╌╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌╌╌╌╌┤
│ #42  ┆ 3         ┆ null  ┆ null      │
├╌╌╌╌╌╌┼╌╌╌╌╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌╌╌╌╌┤
│ #42  ┆ null      ┆ null  ┆ 2         │
└──────┴───────────┴───────┴───────────┘

to this

┌──────┬───────────┬───────┬───────────┐
│ time ┆ instances ┆ rects ┆ positions │
│ ---  ┆ ---       ┆ ---   ┆ ---       │
│ str  ┆ u64       ┆ u64   ┆ u64       │
╞══════╪═══════════╪═══════╪═══════════╡
│ #42  ┆ 3         ┆ 1     ┆ 2         │
└──────┴───────────┴───────┴───────────┘

TODO:

• decide when (on write? on read? both? neither?)
• implement
• test

[teh-cmc]

Doesn't make much sense in the new chunk based model.

It does make sense to do in the range cache, but it is in fact already done there.