[DB-26-3] Open scavenged chunks quickly. Build midpoints later on demand

[timothycoleman]

Changed: Open the database faster by building the midpoints for scavenged chunks later on demand

Otherwise opening a database with thousands scavenged chunks can become too slow

[linear[bot]]

DB-733 Open scavenged chunks more quickly

[sakno]

I have no questions about functional aspects of the proposed changes. From non-functional point of view, there are two improvements possible:

• If lock contention is rare (this is mentioned in the comments inside of the code), the same can be done in lock-free fashion
• RequestCaching might start calculation of midpoints in the background task for each chunk (minimal changes are required) and synchronize at its result only on demand.

[shaan1337]

On my machine, basic performance tests show at least a 10x improvement in the opening speed of scavenged chunks at startup. Previously, it was taking roughly 40ms, now it takes around 1-4ms !