docs: production operations guide for SSE, replay, and backpressure

[catgoose]

Summary

Add an operator-focused guide for running Tavern-backed SSE systems in production.

Why

The library semantics are getting mature, but production success also depends on operational choices:

• proxy/load balancer buffering and timeouts
• replay retention windows
• subscriber pressure and eviction behavior
• observability around disconnects, degradation, and recovery
• sizing/fanout expectations

These concerns are central to the real value proposition of honest live delivery.

Scope

Document:

• proxy/server settings that matter for SSE
• retention/replay tradeoffs
• interpreting backpressure and degradation signals
• suggested metrics/logging
• operational failure modes and expected client behavior

Non-goals

• promising universal deployment recipes for every platform
• replacing app-specific monitoring/alerting

Deliverables

• production guide doc
• README pointers to it
• examples tied back to Tavern control events and delivery states

Acceptance criteria

• a team adopting Tavern has a concrete starting point for production deployment and troubleshooting
• the doc reflects the current delivery semantics rather than generic SSE advice