chore: when i created systimestamp() i noticed that in a single query adjacent records were 7us apart! This was totally atrocious performance that warranted a review. After this refactoring we do three records within 1us. All of this to do with CPU cache cohesiveness.

chore: improved hash function distribution to reduce collisions in maps, that use power-of-two masks. There is an observable performance improvement across the board.

fix(griffin): validating column name in timestamp() clause. Simplified use of timestamp by minimizing static SQL checks in favour of dynamic checks. This would enable SQL functions come up with their own timestamp columns and SQL error checking still working.

fix: utf8 encoding is not supported on windows, yet we used utf8 file and directory names. While it worked for QuestDB, as in we could read what we wrote, to everything and everyone else file system looked like rubbish. I fixed that now, big rewrite of windows filesystem bridge.

fix(test): increasing test wait timeout due to defaulting commit to ASYNC in tests. This makes test runs for longer on CircleCI

fix(lineproto): support nano timestamp precision by default; support boolean values such as T, F, True, TRUE, False, FALSE

fix(griffin): asof joins pre-calculate size of cursor, which saves a lot of time in cases where entire dataset is not needed.

chore(cutlass): refactored LineProtoReceiver to have generic and recvmm() code subclass same base class. Additionally LineProtoReceiver can run on either worker pool, or its own thread for best performance