Implement caching system

[albireox]

This implements caching for Valis queries.

Uses fastapi-cache to optionally cache Valis routes. Two backends are supported, Redis and memcached, which can be toggled using a new setting cache_backend (defaults to redis but memcached does not require ).

I added cache decorators to all the suggested routes in #69 plus some of the /query/list ones except the main query. The library only supports caching GET requests.If this is a dealbreaker we can consider something different for those routes.

On Redis, the key with the cached values is of the form fastapi-cache:valis-cache:get:/query/list/cartons:[] which includes the route and the query params used.

Closes #69

[albireox]

One thing that it doesn't seem to be implemented in fastapi-cache is the option to temporarily (or for easier development) disable caching. The implementation of a backend in fastapi-cache is relatively simple

class Backend(abc.ABC):
    @abc.abstractmethod
    async def get_with_ttl(self, key: str) -> Tuple[int, Optional[bytes]]:
        raise NotImplementedError

    @abc.abstractmethod
    async def get(self, key: str) -> Optional[bytes]:
        raise NotImplementedError

    @abc.abstractmethod
    async def set(self, key: str, value: bytes, expire: Optional[int] = None) -> None:
        raise NotImplementedError

    @abc.abstractmethod
    async def clear(self, namespace: Optional[str] = None, key: Optional[str] = None) -> int:
        raise NotImplementedError

It's probably not super hard to create a null backend that doesn't do anything and always executes the route.

[albireox]

I think this is now ready for review. I've made a few changes:

• We are now using an slightly custom version of the fastapi-cache cache decorator. This one supports POST routes and hashes the contents of the request body. With this we can decorate the main query route. A caveat is that cached routes cannot return an iterator or they will return an empty list in the first request (when they don't hit the cache). I think this is difficult to avoid the way that the decorator works. But I'm also unsure that the generator does much in these cases since the route does need to return the full data payload in one go (unless something like streaming is used).
• I increased the default cache time to 6 months, not sure if we want more or less.
• I replaced the memcached backend (I realised this does require installing software) with an in-memory backend that does not require anything external.
• I added a null backend that does no caching.

[havok2063]

This looks good. I'll pull and test it out first before approving. Is there a way we can explicitly trigger the cache to clear, and wipe the redis or in-memory cache?

[albireox]

To allow clearing the cache I've added a function valis.cache.clear_redis_cache(). If called without arguments it will clear all the keys under fastapi-cache:valis-*.