BrowserEnv CUA Mode Sandbox Teardown

[cdreetz]

When doing training with cua mode BrowserEnv on webvoyager due to some errors it resulted in creating 4k sandboxes I'm assuming bc retrying logic without teardown of past sandbox.

Need to ensure there is a fixed amount of sandbox where no sandbox creation can be attempted unless all past sandboxes are confirmed destroyed. Pool or queue or locks or something idk rn I'll look at it

cc @filip-michalsky