fix(postgresql-proxy): prevent SSL COPY stalls by draining pending SSL buffer

[nik-localstack]

Motivation

Fixes UNC-472: Postgres proxy could hang for sslmode=require sessions running large psql -f / COPY-heavy workloads.

Root cause

selector.select() operates at the TCP layer and is unaware of data already decrypted and buffered inside the SSL layer. After an initial recv(), buffered SSL bytes would not be seen by the next select() call, stalling large transfers.

Changes (postgresql_proxy/proxy.py)

• After each recv() in service_connection(), drain any remaining SSL-buffered bytes using sock.pending() (per Python SSL non-blocking docs).
• Add explicit setblocking(True) on accepted sockets before SSL negotiation. On macOS/BSD, accepted sockets inherit O_NONBLOCK from the listening socket (unlike Linux), causing intermittent BlockingIOError on the MSG_PEEK recv ~1/10 times during SSL startup.

Expected impact

Prevents SSL COPY-session stalls for large dump-like workloads without changing the proxy's overall non-blocking architecture.

[cloutierMat]

Thanks @nik-localstack for the working on that fix and reducing the PR size. I will get to reviewing the stacked PR soon. In the meantime this is GTG. Let's push and publish! 🤘