v4.6.0: Fix race conditions and add persistent failure escalation

Fixes a race condition where the watchdog and boot-fix tasks could interfere
with Claude's VM initialization, and adds escalation logic for unrecoverable
HCS failures.

• Startup grace period now covers ALL watchdog checks including log scanning
  (was only applied to heuristic checks); increased from 90s to 180s
• Persistent failure backoff: after 3 failed fix attempts in 30 minutes,
  the watchdog stops retrying and notifies the user to perform a manual
  Hyper-V nuclear reset (DISM disable/enable)
• Fix script mutual exclusion via global mutex prevents concurrent runs
  from the watchdog, boot-fix task, and manual invocation
• HCS JSON corruption detection (0xC037010D) with specific nuclear reset
  guidance instead of pointless vmcompute restarts
• Watchdog delayed 120s at logon (was immediate)
• Boot-fix delayed 180s at logon (was 90s)

Root cause: On 2026-03-08, HCS JSON corruption (0xC037010D) caused
HRESULT 0x800700DE at VM construct. The watchdog's rapid vmcompute
cycling made the corruption worse. Required a full Hyper-V nuclear
reset to resolve.

Full Changelog: Update4...Update5