fix(client): make retryable stream buffering abort-aware

[Utkarshpandey0001]

Hi Rivet team,

This PR fixes #5800, where ActorHandle.fetch() could remain pending indefinitely when a streaming request body was cancelled while being buffered for lifecycle retries.

Dynamic actor handles created through get() and getOrCreate() support retrying requests during actor lifecycle transitions. Because a ReadableStream can normally only be consumed once, RivetKit buffers the stream before attempting the request. That buffering previously ignored the request’s AbortSignal.

As a result, an already-aborted request—or a request aborted while buffering—would continue waiting for the stream to close. For long-running or open streams, the request would never settle and would never be dispatched to the actor.

What changed

• Check the abort signal before reading a streaming request body.
• Observe the signal throughout the buffering operation.
• Cancel the active stream reader when the request is aborted.
• Preserve and return the original abort reason to the caller.
• Release the stream lock and remove the abort listener during cleanup.
• Keep the existing retry behavior unchanged for successfully completed streams.

Test coverage

Regression tests were added for:

• An already-aborted streaming request made through getOrCreate().
• A request aborted while its stream is being buffered through get().
• Preserving the original abort reason.
• Cancelling and unlocking the source stream.
• Ensuring no HTTP request is dispatched after cancellation.
• Confirming existing streaming-body retry behavior continues to work.

Verification

• Focused fetch-retry suite: 11 tests passed.
• TypeScript type checking passed.
• Biome formatting and lint checks passed.

Closes #5800