Integrate vfs2

[codecov[bot]]

Codecov Report

Attention: Patch coverage is 71.08434% with 168 lines in your changes missing coverage. Please review.

Project coverage is 77.35%. Comparing base (6633bd8) to head (86ef5cd). Report is 29 commits behind head on master.

Files	Patch %	Lines
src/fsm.c	56.25%	34 Missing and 15 partials :warning:
src/server.c	61.34%	33 Missing and 13 partials :warning:
src/vfs2.c	74.78%	10 Missing and 19 partials :warning:
src/raft/replication.c	58.06%	7 Missing and 6 partials :warning:
src/leader.c	61.90%	5 Missing and 3 partials :warning:
src/raft/uv.c	77.27%	2 Missing and 3 partials :warning:
test/raft/integration/test_uv_load.c	84.61%	4 Missing :warning:
src/utils.h	57.14%	0 Missing and 3 partials :warning:
src/raft/uv_metadata.c	75.00%	0 Missing and 2 partials :warning:
src/raft/uv_segment.c	92.00%	0 Missing and 2 partials :warning:
... and 5 more

Additional details and impacted files

```diff @@ Coverage Diff @@ ## master #661 +/- ## ========================================== - Coverage 77.41% 77.35% -0.07% ========================================== Files 196 196 Lines 27269 27377 +108 Branches 5455 5519 +64 ========================================== + Hits 21111 21178 +67 + Misses 4297 4286 -11 - Partials 1861 1913 +52 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

[cole-miller]

This PR does a few different things (sorry):

• It removes --enable-dqlite-next from the build system and all mention of DQLITE_NEXT in the source code. All the code for the new disk-based configuration is now compiled in all build configurations, except for uses of new raft APIs that aren't available when linking against an old, external libraft.
• It introduces new codepaths that use vfs2 instead of the old VFS, including calling functions that have no counterpart in the old setup, like vfs2_unapply. I've repurposed the existing dqlite_node_enable_disk_mode option to switch between the old in-memory VFS and vfs2; the existing on-disk configuration of the old VFS is no longer reachable after this PR.
• It introduces a run-time configuration option for the raft_uv I/O implementation, raft_uv_set_format_version, that toggles between the backwards-compatible format for the persistent log and the new format (originally introduced and gated behind DQLITE_NEXT in #639).
• Includes various fixes for vfs2 that were added in the course of getting our test suite to pass when running in "disk mode". Note that the changes included in this initial pass are definitely not the final word---for example, as discussed elsewhere, vfs2_pseudo_read_begin needs to be rewritten and the way the WAL-index header is handled needs to be redone, also some of the fixes in this PR are more like band-aids and need to be redone more carefully.
• Finally, reworks the initialization logic in server.c, moving things around so that we no longer leak resources if you create a dqlite_node and then destroy it without running it first (a longstanding issue, see comments in this file on master). I got sucked into doing this while working on the extra initialization required to set up vfs2, although it wasn't strictly necessary to carry it as far as I have done here.

[cole-miller]

Closing in favor of more focused PRs.