Add query_cache option to selectively bypass query cache

[rbino]

Highly dynamic queries with many different combinations of joins can generate unique cache keys on every execution, causing cache bloat while providing little performance benefit. This option allows selectively disabling Ecto's internal query cache for such cases, preventing cache pollution while still maintaining normal caching behavior for typical queries.

When set to false, the query bypasses cache lookup, cache insertion, and cache update callbacks, going directly to query normalization and adapter preparation on each execution.

[greg-rychlewski]

Thanks :)

@josevalim I could think of a couple small follow ups that might be good. If you agree, I wouldn't mind taking a stab at these:

1. Make it so this option takes priority over our automatic cache disables. So basically enabling the opposite direction this PR is addressing. For example we disable the caching of values because we got some complaints about it not hitting the cache often enough. But there will be cases where people can be sure it will hit the cache.
2. I think it should be possible to cleanly move this option deeper into the planner so that we don't generate the key at all if the option is set to false