feat: add opt-in TCP_NODELAY support for low-latency trading

[vjsingh1984]

Problem

Trading systems using this library need to minimize latency on order submissions. Currently, the TcpStream connections use Nagle's algorithm (the default), which can buffer small writes (order messages ~100-200 bytes) for up to 40ms before sending them.

For algorithmic trading, this adds unnecessary latency to order routing — a critical path where milliseconds matter.

Proposed Solution

Add opt-in TCP_NODELAY support via:

1. Environment variable IBAPI_TCP_NODELAY=1 to disable Nagle's algorithm
2. Apply to all 4 connection paths:
   • TcpSocket::new() (sync initial connection) — transport/sync.rs:704
   • TcpSocket::reconnect() (sync reconnect) — transport/sync.rs:719
   • AsyncConnection::connect_with_callback() (async initial connection) — connection/async.rs:42
   • AsyncConnection::reconnect() (async reconnect) — connection/async.rs:91

Backward Compatibility

• Default behavior: unchanged (Nagle enabled)
• Opt-in: Set IBAPI_TCP_NODELAY=1 before calling Client::connect()
• No API changes required
• Zero breaking changes

Performance Impact

When enabled:

• Eliminates up to 40ms delay on small writes (order submissions, market data requests)
• For trading systems with low message rates (~1-10/sec), bandwidth overhead is negligible
• Significantly reduces order routing latency

Implementation

I have a fork with this change ready to submit as a PR. The implementation:

• Adds 4 socket.set_nodelay(true)? calls, one per connection path
• All 100 existing unit tests pass
• Environment variable is checked at connection time only
• ~18 lines of code total

Would you be open to this feature? I can open a PR if this aligns with the project's goals.

[wboayue]

That's a useful feature. Added in #394