fix(providers): auto-retry or recover from mid-stream connection aborts (wsarecv / connection reset)

[euxaristia]

Summary

During active streaming, transient socket interruptions such as Windows wsarecv: An established connection was aborted by the software in your host machine (WSAECONNABORTED) or read: connection reset by peer immediately abort the active turn with Run cancelled., requiring the user to manually type continue.

Context

When an established stream connection drops mid-generation before any tool calls or side effects are dispatched:

• The provider stream decoder encounters a network read error.
• The turn immediately terminates and surfaces the error to the composer.
• While pre-send connection failures are retried (Auto-retry transient network errors (TLS timeout, conn reset) instead of requiring manual "continue" #447), mid-stream socket aborts currently require manual user intervention even when no side effects were executed.

Proposed Solution

1. Safe Mid-Stream Retry: When a stream drops before any tool execution begins, auto-retry the request with exponential backoff before failing the turn.
2. Transient Error Classification: Classify WSAECONNABORTED, ECONNRESET, and unexpected stream EOFs as retryable transport errors when the current turn has executed no tool calls.