fix(inject_fault): detect QEMU monitor prompt across split recv() chunks , minor bugs too

[Anshul1596]

What

Fixes a bug where send_cmd() only checked the latest recv() chunk for the (qemu) prompt. If the prompt was split across multiple reads, every command waited for a timeout, slowing down fault injection—especially ring_flood.

Changes

• Fix send_cmd() to check the accumulated response for (qemu).
• Batch all ring_flood NMI triggers into a single sendall().
• Use os.urandom(16) in fault_nvs_corrupt() for faster random bytes.
• Ensure the monitor socket is always closed with try/finally.

Testing

Tested all fault types against a local QEMU instance. Verified behavior is unchanged, with ring_flood completing noticeably faster.