asyncnet ssl overhaul

[nitely]

Fixes #24895

• Remove all bio handling
• Remove all sendPendingSslData which only seems to make things work by chance
• Wrap the client socket on acceptAddr (std/net does this)
• Do the SSL handshake on accept (std/net does this)

The only concern is if addWrite/addRead works well on Windows.

[Araq]

What was the insight gained here?

[nitely]

The official guide for client and server use the SSL_* API family, which underneath use the BIO_* APIs. BIO_* allows to handle the I/O (send/recv) in application code. SSL_* handles the I/O internally, and returns WANT_READ/WANT_WRITE (note a ssl_read may want to write and ssl_write may want to read, because of renegotiation) so the socket can be awaited to become readable/writeable and repeat the operation.

Nothing I found in the docs hints one should mix BIO_* I/O with SSL_* like we did; although that does not mean this is not valid, it does not look like the way to do it. If using BIO_ were required, I think the way would be copying what SSL_[READ, WRITE, CONNECT, ACCEPT] do into the application code, and get rid of all the SSL_*, which does not seem future-proof/wise either. The only example I found of using BIO is this but it's using a blocking socket, which takes care of things like renegotiation.

Looking at std/net, there is almost a 1:1 map to std/asyncnet after these changes wrt error handling.

[github-actions]

Thanks for your hard work on this PR!
The lines below are statistics of the Nim compiler built from 8518cf0

Hint: mm: orc; opt: speed; options: -d:release
179420 lines; 9.178s; 651.219MiB peakmem