Always cache the full result

[morozov]

Q	A
Type	bugfix, improvement
BC Break	no

Fixes #4178. Closes #4189.

1. Cache the whole result right after it's available regardless of how it's fetched and whether it's fetched at all. The original implementation of caching only after the result is fully fetched optimizes for the edge case and shouldn't exist.
2. Serve the result from an array result which is the same as if the result was retrieved from the cache.
3. Move the result caching logic from CachingResult to the wrapper connection. This way, we can remove the redundant call to the cache.
4. Remove CachingResult as no longer used.
5. Remove testDontFinishNoCache() as no longer valid.
6. Rename ResultCacheTest since it covers the whole feature, not a specific class.

[greg0ire]

I think docs/en/reference/caching.rst will need to be changed, since users no longer need to explicitly read the entire result set from the database. We should instead warn them that using the caching feature does that implicitly. Maybe we should also document it as a minor BC break in the upgrade guide?

[morozov]

I think docs/en/reference/caching.rst will need to be changed, since users no longer need to explicitly read the entire result set from the database.

Thanks! Updated.

We should instead warn them that using the caching feature does that implicitly.

Should we? IMO the current implementation is the most reasonable and doesn't need documenting technical details.

Maybe we should also document it as a minor BC break in the upgrade guide?

What is the BC break?

[greg0ire]

What is the BC break?

The result set is now implicitly read entirely. If somebody uses this API to read the first row of a query that returns many results, they might have memory issues. But As I write this lines, I realize this is too far-fetched 😅 . Ignore me.