feat: Programs edit results carry rw2's historyWarning

[DanielKow]

Summary

When an rw2 program edit lands but its history commit fails, rw2 answers 200 with the X-History-Not-Recorded: true header and a historyWarning body field (rw2 ProgramEditingEndpoints.cs:416-427 for add/edit/remove/move, :296-302 for capture). RocketWelder.SDK.Http dropped both, so every consumer (the MCP server first) reported a plain success.

The client now reads historyWarning from the body. rw2 sets that field on every landed-edit body whenever it sets the header, so the body is the single source and the header is not read separately.

• BlockEditResult (add/edit) and CapturePointResult (capture) gain string? HistoryWarning (an init property; this change is additive).
• Breaking: RemoveBlockAsync / MoveBlockAsync now return EtagEditResult(ProgramEtag Etag, string? HistoryWarning) instead of a bare ProgramEtag. Release this as a major version.
  • Callers must read .Etag.
  • The only other implementer found in the org is rw2's test fake StationManagement.Tests/ProgramAuthoringTests.cs:91 (RecordingPrograms : IProgramsApi, the Remove/Move members at :135/:137, on the epic-97 branch). It needs a two-line change when rw2 bumps past this version. rw2 pins 2.27.0 / 2.28.1 today, so nothing breaks until then.

Tests

• ProgramsHistoryWarningTests (new, written first).
  • For add, edit, remove, move and capture: a stubbed 200 reply with X-History-Not-Recorded and a historyWarning yields that text verbatim, and a reply without them yields null.
  • RED before the fix: 14 compile errors, because the result types had no HistoryWarning and remove/move returned ProgramEtag.
• GREEN: RocketWelder.SDK.Http.Tests passes 74/74, and the full RocketWelder.SDK.sln build succeeds.
• Mutation check: dropping the field in ReadEtagAsync (passing null) fails exactly the remove and move warning tests.

Release

Not released yet. release.sh runs only on the user's go-ahead. The companion mcp PR pins the released version.

[DanielKow]

Epic 097 / F11-11b review of 2424220.

Sound:

• HistoryWarning is taken from the response BODY only. It is deserialised into BlockEditResult / CapturePointResult, and EtagEditResult is built from EtagResponse.
• Remove and Move now return EtagEditResult. That is a breaking change to the public API, and it matches the 3.0.0 packed version.
• No AI trailers.

Breaks (detached worktree; control 36/36; each reverted by its inverse, diff 0):

• new EtagEditResult(etag, null) ⇒ RED MoveBlockAsync_/RemoveBlockAsync_Should_Carry_The_HistoryWarning_Of_A_Not_Recorded_Edit.
• BlockEditResult.HistoryWarning => null ⇒ RED AddBlockAsync_/EditBlockAsync_Should_Carry_The_HistoryWarning_Of_A_Not_Recorded_Edit.

COULD: every "not recorded" fixture sends the header AND the body. No test pins that the body alone drives the field; for example, a body warning with no header should still carry it, and a header with no body should give null.

— Reviewer