[pycurl] Fill in missing/Incomplete annotations

[Psychosoc1al]

This PR fills in missing/incomplete type annotations for pycurl package.

Reasoning

Missing annotations:

• Curl.setopt() (docs, source code): Since this method accepts a wide variety of types, I was a bit scared of a huge annotation. It would include all kinds of objects, ranging from int, str and file-like objects to multiple signatures of Callable and something like [(str | bytes, (str, bytes, (int, str | bytes)))] (for legacy, but supported HTTPPOST). For now, I have annotated it with Any | None. I am open to suggestions on how to structure this more strictly.
• CurlMulti.setopt() (docs, source code): This method is much more restricted than previous one. I went through documentation and source code and did a small brute-force over all available 12 options for this method to ensure.
• CurlShare.setopt() (docs, source code): Similar to the above, but only limited to two options that only accept int. Also ensured with a simple brute-force check.
• CurlMulti.assign() (docs, source code): The value parameter is Any | None by design. None is used to unassign, and other Python objectscan be passed to store their reference so they can be retrieved later in a callback without any processing by curl itself (works roughly like this: store "socket: object" relation -> *something happens on that socket* -> callback called -> object related to that socket is passed to callback).

Incomplete annotations:

• version_info() (docs, source code): These Incomplete fields use the same underlying string conversion function (vi_str()) as other string-returning fields in the structure. However, these two specific values can also be None, which seems to match libcurl documentation.
• CurlMulti.perform() (docs, source code): The "status" returned is just an int, similar to other functions: libcurl docs.
• CurlMulti.fdset() (docs, source code): The C-level file descriptor sets (fd_set) are translated to Python list[int], here source code is pretty straightforward.
• CurlMulti.info_read() (docs, source code): The return structure is a bit more complex, but described in details in documentation

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

tornado (https://github.com/tornadoweb/tornado)
+ tornado/curl_httpclient.py:55: error: Argument 2 to "setopt" of "CurlMulti" has incompatible type "Callable[[int, int, Any, bytes], None]"; expected "bool | int | list[str | bytes] | tuple[str | bytes, ...] | Callable[[int], Literal[-1, 0] | None] | Callable[[int, int, CurlMulti, Any | None], Literal[-1, 0] | None] | None"  [arg-type]

[Psychosoc1al]

The first error mentioned in #15919 (comment) was indeed a bug on my end: I had mixed up the unsupported SOCKOPTFUNCTION with the supported SOCKETFUNCTION. I have corrected this, and the annotation now matches the documentation.

However, the failure in #15919 (comment) appears to be a downstream issue. It stems from a loosely annotated, unused multi: Any parameter in Tornado's implementation (tornado/curl_httpclient.py:93), which is now conflicting with the tightened types in pycurl.

How should I proceed with this conflict?