feat(hamilton): add configurable STAR chatterbox state

[VivienP]

Summary

Implements configurable per-instance STAR chatterbox query state, related to #934. Parts of that issue are stale relative to current main: several listed methods are already tracker-backed, LLD-latched, or firmware-path commands rather than hardcoded chatterbox replies.

This PR only configures the remaining category-A hardcoded simulation readings.

Configurable fields

• iswap_initialization_status → request_iswap_initialization_status() (default True)
• channel_z_positions → request_z_pos_channel_n() (default 285.0 mm)
• dispensing_drive_positions → channel_dispensing_drive_request_position() (default 0.0)

Why per-instance typed state

A mock_response parameter on every query would diverge chatterbox signatures from STARBackend and would not give a stable reading across a multi-step protocol. STARChatterboxState is bound per backend instance, copied into tuples, and replaced wholesale through backend.chatterbox_state.

Backward compatibility

STARChatterboxBackend() keeps the current unconfigured replies. STARBackend is unchanged. channel_dispensing_drive_request_position(..., simulated_value=None) remains source-compatible for existing callers; an explicit simulated_value=0.0 is distinct from an omitted argument.

One source of truth (excluded)

• Tip presence, 96-head tip presence, and tip length stay on TipTracker
• Last LLD heights stay latched by simulated LLD
• Parked flags stay on existing backend latch state
• Machine / extended / iSWAP factory configuration stay on their existing objects
• Carrier-presence and CoRe resource-existence commands still follow the firmware path

Non-goals

This is simulated query state, not physical evidence. It is not a physics simulator and not a fault-injection framework. Motion commands do not update these readings.

Test plan

• STAR tests
• generic chatterbox tests
• STARChatterboxState tests (defaults, isolation, copies, invalid lengths, index checks, replacement, setup() preservation, simulated_value=0.0 vs omitted, tracker/LLD ownership)
• LiquidHandler tests
• TipTracker tests
• ruff check pylabrobot
• ruff format --check pylabrobot
• mypy pylabrobot
• git diff --check

Local Sphinx dummy -W on Windows reports 11 pre-existing autosummary stub warnings from case-insensitive plate alias collisions (BioRad_* vs biorad_*, etc.). simulation.md itself built without warnings. CI docs-check runs on Ubuntu.