fix: fail task and release machine on unexpected guest errors

[itamarga]

Problem

When the guest agent returns an error during analysis — e.g. an HTTP 500 from /store while the host uploads the sample — the resulting requests.HTTPError escapes CAPE's error handling and orphans the task: the VM is left running, the machine stays locked, and the task sits in running forever. The analysis timeout never fires because it only applies once wait_for_completion() has begun, which this never reaches.

We hit this in production: three submissions whose sample filename collided with a pre-existing path in the guest's %TEMP% each 500'd on upload and left all three machines locked for 3 days until manually cleaned up.

Root cause (lib/cuckoo/core/analysis_manager.py)

• machine_running() — the stop_machine() + release() cleanup sat after the try/except, so it only ran on the no-error path. The except only catches (CuckooMachineError, CuckooGuestCriticalTimeout); any other exception through yield skipped cleanup entirely → VM left running + machine locked.
• launch_analysis() — only finalized task status for CuckooDeadMachine (→ pending) or clean completion (→ completed). Any other exception propagated out with the task left running.

Fix

• machine_running(): always stop/release the machine via a finally block (extracted into _stop_and_release_machine()), skipping only the dead-machine case which already removes the machine.
• launch_analysis(): on any unexpected exception, mark the task failed_analysis and unlock the machine.

Tests

Added regression tests in tests/test_analysis_manager.py:

• test_machine_running_releases_machine_on_unexpected_error
• test_launch_analysis_marks_failed_on_unexpected_error

tests/test_analysis_manager.py: 19 passed, 5 skipped (pre-existing data-file skips).