Change python detectors cache implementation

[edena-legit]

Currently, the Python client that the detectors use cache the whole response, which leads to OOM when the whl files are big.
First, I think the cache should be in the PythonResolver level and not in the client and cache Uri to HashSet of the dependencies.
Also, I think the cache should be changed to IMemoryCache instead of an actual MemoryCache instance.

AB#2098100

[melotic]

I'll take a look into this, but I believe that the new Simple PyPI detector does not cache the wheel file.

[edena-legit]

It does, line 117 in SimplePyPiClient.cs.
Either way, there's an issue with the way the request is sent. We are using HttpClient.SendAsync, which throws OutOfMemory. Instead, we should use HttpClient.GetStreamAsync.

https://stackoverflow.com/questions/21169370/httpclient-getasync-out-of-memory-exception