Replication metadata rework. Migrate from crsqlite 0.17 to 0.18

[gorbak25]

Closes #532, #366, #188

• After migrating over the full production global corrosion db + pruning tombstone_pks it decreases storage by 3 times without Explore active GC #533
• The existing synthetic benchmarks we have shown a decent performance improvement even without future opts like Add bulk change processing interface to crsqlite #562

Updates only(100% updates):
| Batch | V1 | V2 packed | Change |
| --- | --- | --- | --- |
| 1 | 4.04 ms | 4.31 ms | +6.7% |
| 2 | 4.56 ms | 3.23 ms | −29.1% |
| 5 | 4.39 ms | 3.81 ms | −13.2% |
| 10 | 8.12 ms | 4.31 ms | −46.9% |
| 25 | 8.49 ms | 7.15 ms | −15.9% |
| 50 | 17.39 ms | 12.67 ms | −27.1% |
| 100 | 38.66 ms | 27.76 ms | −28.2% |

Mixed workloads(10% inserts, 80% updates, 5% deletes, and 5% true resurrections):
| Batch | V1 | V2 packed | Change |
| --- | --- | --- | --- |
| 1 | 4.47 ms | 4.78 ms | +6.9% |
| 2 | 5.83 ms | 3.45 ms | −40.8% |
| 5 | 4.24 ms | 4.36 ms | +2.7% |
| 10 | 5.67 ms | 5.18 ms | −8.6% |
| 25 | 9.71 ms | 9.18 ms | −5.5% |
| 50 | 17.78 ms | 16.07 ms | −9.6% |
| 100 | 37.52 ms | 32.57 ms | −13.2% |

Due to the scope of the PR it ought to cause issues. So here's a hall of shame of incidents caused while trying to ship this:

• https://status.flyio.net/incidents/s9dzwg3d4bs7