RobotError is waldoctl's; waldoctl pin -> v0.11.1

[Jepson2k]

A frontend represents a refused command the same way whichever backend raised it, so the refusal type is the contract's: parol6.RobotError is now waldoctl.RobotError. Same six fields, same wire list in both directions, and an exception rather than a dataclass, so a client can raise it as-is. make_error, the catalog and extract_robot_error are unchanged.

One call site changed with it: the controller re-attributed a tool error with dataclasses.replace, which an exception cannot take; it rebuilds from the wire fields instead. No test reaches that path — that gap was there before and is still there.

Verified against waldoctl v0.11.1 plus Jepson2k/waldoctl#19 (which adds to_wire and makes the class copy/pickle-safe — parol6's state snapshot deepcopies it): 194 unit and 79 integration tests pass. Neither parol6 nor Waldo Commander used the sync-wrapper classes waldoctl's v0.11.0 removed and v0.11.1 restored; make_sync_tool is unchanged in signature.

The pin moves from v0.7.0 to v0.11.1; it should move to v0.11.2 once waldoctl#19 is tagged.

🤖 Generated with Claude Code

https://claude.ai/code/session_014Bo12kumRx9PHnY9bL8qgn

[Jepson2k]

Superseded: the branch is renamed to feat/thin-python-binding so the cross-repo CI branch-matching resolves it (waldoctl <- par6 <- Waldo-Commander all use that name). Same commits, continued in the new PR.