fix: sweep-cleanup lingering sandboxes after each benchmark mode

[HeyGarrison]

Summary

• After each provider finishes a benchmark mode (sequential, staggered, burst), call compute.sandbox.list() and destroy any sandboxes still alive
• Logs how many lingering sandboxes were found and whether destroys succeeded
• Prevents resource accumulation across modes — daytona was hitting quota exceeded on burst (51/100 failures) because sandboxes from earlier modes were still alive

Problem

We run 3 modes per provider: sequential (1 sandbox), staggered (100 sandboxes), burst (100 sandboxes) = ~201 total. The per-iteration finally blocks call destroy(), but:

• Destroy can time out (15s) and we move on
• The provider's delete API may be eventually consistent
• @computesdk/daytona silently drops autoStopInterval/autoDeleteInterval from sandbox options, so sandboxes aren't ephemeral

By burst mode, dozens of sandboxes from earlier modes are still lingering, eating quota.

Fix

Added cleanupAllSandboxes(compute, providerName) that runs after each mode per provider:

1. Calls compute.sandbox.list() (30s timeout)
2. Destroys everything found via Promise.allSettled (15s timeout each)
3. Logs results so we have visibility into what's being cleaned up