fix(liquid-handling): report committed tip state from chatterbox sensors

[VivienP]

Summary

• Separate logical/pending tip state from simulated physical/committed tip state.
• Add TipTracker.has_committed_tip and get_committed_tip().
• Make Chatterbox simulated hardware queries use committed state:
  • generic request_tip_presence()
  • STAR request_tip_presence()
  • STAR 96-head tip presence
  • STAR simulated tip-length query
• Fix LiquidHandler recovery in simulation so a failed pickup/drop does not observe its own pending tracker mutation and incorrectly commit it.
• Add focused regression coverage for failed pickup/drop recovery, ChannelizedError, pending-vs-committed sensor behavior, and STAR simulated queries.

This remains independent of #1206 and is compatible with #1206 changing get_tip() to expose pending state.

Hardware UNKNOWN/uncertain state, generic retry logic, partial physical success modeling, and fault-injection infrastructure remain out of scope.

Test plan

• TipTracker committed-state tests
• Chatterbox recovery tests
• LiquidHandler recovery tests
• STAR Chatterbox tip presence / tip length / 96-head tests
• Ruff check / format

Compatibility was explicitly tested with #1206 semantics (get_tip() reading pending state).

Combined validation with #1206 on current upstream/main:

• 236 tests passed
• 101 subtests passed
• ruff check pylabrobot passed
• ruff format --check pylabrobot passed