Fix SandboxGetLogs in Typescript client

[saltzm]

Implement retry backoff for sandbox log fetches, make stdout/stderr lazy, and ensure cancelable iterators stop consumption to improve resource management and responsiveness.

The previous implementation eagerly started log streams (SandboxGetLogs) upon Sandbox object creation, even if stdout or stderr were never accessed. This PR changes stdout and stderr to be lazy-initialized, only starting the underlying gRPC stream when first accessed. Additionally, it adds exponential backoff to SandboxGetLogs retries to prevent hammering the server on transient errors and modifies stream cancellation to immediately stop consumption, preventing unnecessary resource usage. This also enables a previously skipped test that relied on lazy log streaming behavior.

---

Slack Thread

 

[cursor]

Cursor Agent can help with this pull request. Just @cursor in comments and I'll start working on changes in this branch.
_{Learn more about Cursor Agents}

[cursor]

PR Summary

Medium Risk
Changes sandbox log streaming lifecycle (lazy init, cancellation, retry backoff), which can subtly affect stdout/stderr consumption timing and error behavior; new tests reduce but don’t eliminate behavioral risk across runtimes.

Overview
Improves SandboxGetLogs handling in the TS client by making Sandbox.stdout/Sandbox.stderr lazily initialized (log streams are no longer started during Sandbox construction) and wiring cancellation via AbortController/AbortSignal so reader cancellation stops the underlying gRPC stream.

Adds short exponential backoff and retry counter reset behavior to the sandboxGetLogs loop to avoid tight retry hammering on transient errors, and updates streamConsumingIter to propagate cancellation upstream via AsyncIterator.return() instead of draining the iterator. Re-enables the previously skipped experimental docker mock test and adds a focused sandbox_logs.test.ts suite covering laziness, retry/backoff, resume via lastEntryId, error propagation, and cancellation cleanup.

^{Written by Cursor Bugbot for commit bd129b4. This will update automatically on new commits. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}

[ehdr]

This overall looks good to me! Sent some comments over Slack too.

I believe I made the Go implementation lazy as well, but we don't have the backoff for retries, so I'd guess Go could also end up in a tight retry loop? So would be good to look into that as well!