fix: use persistent httpx.Client to eliminate per-call TCP/TLS overhead

[sameer-ahmed]

Problem

httpx.request() (the top-level convenience function) creates a brand new TCP connection and TLS handshake on every single API call. On internal/corporate Redmine servers this adds ~2 minutes of overhead per request, making bulk operations extremely slow.

We measured this empirically:

• Fetch 1 ticket: ~2 min 3 sec
• Update 1 ticket: ~2 min 6 sec
• 10 calls total (5 fetch + 5 update) = ~20 minutes for what should take seconds

Fix

Replace httpx.request() with a module-level httpx.Client instance that reuses the underlying TCP/TLS connection across calls. After the first connection is established, subsequent calls reuse it — reducing latency to milliseconds.

# Before: new connection every call
response = httpx.request(method=..., verify=..., timeout=...)

# After: persistent client, connection reused
_http_client = httpx.Client(timeout=60.0, verify=not REDMINE_DANGEROUSLY_ACCEPT_INVALID_CERTS)
response = _http_client.request(method=...)

Why this happens

httpx.request() is a convenience wrapper that internally does:

with httpx.Client(...) as client:
    return client.request(...)

The with block closes the client (and its connection pool) after every call. The persistent client avoids this entirely.

This is consistent with the httpx docs recommendation — use a Client instance for multiple requests rather than the top-level API.

Impact

• Fixes severe latency on corporate/internal Redmine instances (HTTPS with self-signed certs or slow DNS)
• No behaviour change for public Redmine instances (just faster)
• Zero API surface change