CrateDB's SQLAlchemy dialect now handles the `server_default` when generating
table DDL.

Fix #454.

Co-authored-by: Mathias Fußenegger <[email protected]>

The SQLAlchemy implementation of CrateDB's `OBJECT` type offers indexed
access to the instance's content in form of a dictionary. Thus, it must
not use `cache_ok = True` on its implementation, i.e. this part of the
compiled SQL clause must not be cached.

Tests: Add integration test cases verifying SA's SQL statement caching
Specifically, the special types `OBJECT` and `ARRAY` are of concern here.

SQLAlchemy's `sqltypes.JSON` provides a facade for vendor-specific JSON
types. Since it supports JSON SQL operations, it only works on backends
that have an actual JSON type, which are currently PostgreSQL, MySQL,
SQLite, and Microsoft SQL Server.

This patch starts leveraging the same infrastructure, thus bringing
corresponding interfaces to the CrateDB dialect. The major difference
is that it will not actually do any JSON marshalling, but propagate
corresponding data structures 1:1, because within CrateDB's SQL,
`OBJECT`s do not need to be serialized into JSON strings before
transfer.

This method supports efficient batch inserts using CrateDB's bulk
operations endpoint.

https://crate.io/docs/crate/reference/en/latest/interfaces/http.html#bulk-operations

Co-authored-by: hlcianfagna <[email protected]>

Specifically, outline _two_ concrete considerations for determining
the optimal chunk size, and improve wording.