Improve http.send cache lookup and freshness logic

[anderseknert]

The current approach to cache lookups and freshness checks is to store the headers as sent by the remote server, and each time we check if an entry should be fetched from cache or from the remote, we parse all the relevant date-related headers — Date, Cache-Control (including max-age) and Expires in order to determine if the cache entry is fresh.

Let's instead do this parsing only once — when the cache entry is created, and store the timestamps directly for comparison in the freshness checks. This would improve performance of cache lookups, but it would also nicely abstract away the details of the HTTP spec from the caching logic.

In order to do this, we'll need to extend the test suite for inter query caching in http.send, which surprisingly was almost non-existent before #5298 .. while some e2e tests are added there, we'll probably need another approach to test expiry.

[lukyer]

@anderseknert does this also improve some kind of automatic cache items invalidation when the max-age time is reached? We store a lot (thousands) http.send responses with low TTL (10s). Currently it is unnecessarily hanging in the cache until caching.inter_query_builtin_cache.max_size_bytes is reached (and when it is, items get wiped in FIFO manner instead of respecting the TTL).

Thanks!

[srenatus]

Could you open an issue, please? (Or search for an existing one?) It seems like having a periodic cache-cleaning routine would help you, but let's flesh out the details together.

[lukyer]

Sure, I wasn't sure if this improves the situation somehow or not. I'm not too familiar with your codebase yet.
If you think this does not improve the cache cleaning behaviour, I'm happy to create new feature request issue.

[srenatus]

The idea here was to straighten something out that was a bit quirky in the code. No functional changes involved. Thanks for the issue!

[anderseknert]

@lukyer: @srenatus is right about that. However, this change does pave the way for what you suggest, as this would have been quite expensive before, as we'd have to iterate over each item, parse all the date-related headers, and check for expiry. Now we can just check for expiry directly.