Internal WAL read-write lock is very expensive

[penberg]

We have an internal WAL read-write lock, which is very expensive, especially for single-threaded workloads. We should either move locking to higher layers (e.g. in the SQLite C API) or have some less expensive lockless synchronization mechanism internally.

[LtdJorge]

So, I've been implementing a SeqLock since yesterday which I'm hopeful will solve at least part of the problem, I'll send a PR later today.

One thing @penberg did you use a test harness or benchmark specific to the RwLock when you found the performance problem? I'm asking this, mainly because it would help me a lot in benchmarking my implementation vs what we currently have.