1. Rebase on top of fresh 2.11.0-picodata branch (2.11.0-40-g8277665cf0)
2. Drop two commits related to cracklib vendoring. The original feature
   was abandoned so the cracklib isn't needed anymore.

Denis Smirnov authored 1 year ago and Yaroslav Dynnikov committed 1 year ago

Before current commit, sql waited for ddl operations to be prepared
in the cluster, but not committed or aborted. As a result, when the
user queries the space right after the previous ddl successfully
returned "ok", it was possible to retrieve "no space found" error.

Also fix some fresh clippy warnings

This reverts commit 5a5a1919.

When picodata process forks, the parent and child are connected by
two unix pipes:
- the fuse (used by the child to detect parent's death);
- the ipc channel (used by the parent to get on-exit messages from
  the child to change its state in the finite-state machine);

The problem was that the parent tried to read the message from the
pipe right after the fork (and before the child's exit). Usually
the pipe was empty at that moment and the parent collected serde's
"failed to fill whole buffer" error instead the real message provided
by the child.

Most of them are already moved to the site repo:
https://git.picodata.io/picodata/picodata/docs/-/tree/97c3e35ec/docs

- `clustering.md` -> https://docs.picodata.io/picodata/clustering/
- `glossary.md` -> https://docs.picodata.io/picodata/glossary/
- `cli.md` -> https://docs.picodata.io/picodata/cli/
- `discovery.md` -> https://docs.picodata.io/picodata/discovery/
- `sharding.md` -> /dev/null
- `topology.md` -> https://docs.picodata.io/picodata/deploy/