[Blind-Striker]

📝 Description

What does this PR do?
It moves OpenCode.Sdk.Tests helpers and embedded fixtures out of the project root and into the slice whose code uses them, as testing-style.md §1 already says. It is moves only: no test body, assertion or helper behaviour changes.

• 146 moves: 92 fixture assets and 54 C# files (helpers plus their self-tests). Each was classified at base cde7e254 by its code consumers only. A file stays in root Support/ or Fixtures/ when more than one slice uses it.
• Fixture identity is preserved. Every moved EmbeddedResource keeps its old LogicalName (OpenCode.Sdk.Tests.Fixtures.<old path with dots>) and is included exactly once, so no loader changes. The two oracle scripts (tools/oracles/first-match-union.ts, wildcard-path.ts) read their fixtures from disk and now point at the new folders.
• Forced edits only:
  • namespaces and using directives of moved files and their consumers;
  • one SdkTestSupport.HangGuard qualification;
  • the .editorconfig section paths;
  • the repository-gate path lists;
  • the suppression-ledger snapshot rows (same rows, new paths and sort order);
  • the testing-style.md §3 fixtures sentence, which now names the slice's Fixtures/ folder, or the root one for shared fixtures.
• Plan corrections, each decided by the code-consumer rule:
  • PathCommandShim → Hosting/Launcher/Support/: its only "Harness consumer" was a doc comment.
  • BlockingContent and Gate stay in the root Support/. [FRZ-H2b] per-call request options on the stream and traversal doors; the stream-failure Tag on the marker #398 added a Sessions consumer of BlockingContent (Sessions/SessionClientContractTests.cs:60). That makes it cross-slice, and Gate is reached from both slices through it. InterruptedBufferingContent is still Transport-only and moves.
• Rebased onto [FRZ-H2b] per-call request options on the stream and traversal doors; the stream-failure Tag on the marker #398 (cde7e254). Rename detection carried [FRZ-H2b] per-call request options on the stream and traversal doors; the stream-failure Tag on the marker #398's edits onto the three moved helpers it touched (OwnedToolScenario, SessionLogTranscript, SimulatedSessionTurn); against [FRZ-H2b] per-call request options on the stream and traversal doors; the stream-failure Tag on the marker #398's versions they differ only in namespace and using lines. The nine overlapping consumer files differ from master only in using lines. One import hunk from [RT-392] perf: pool downlevel response bodies between 1 and 16 MiB #397 was resolved by hand earlier.

Part of TST-E7 (step 3a, the SDK half; the Tools half follows as a separate PR).

🔄 Type of Change

• 🧪 Test improvements

🎯 Protocol & API Surface

• No change to the public API surface
• Nothing under external/, spec/ or src/ was touched

🧪 Testing

Clone baselines. The second commit carries only the tests baseline's own delta: 16 using-header fingerprints added and 4 removed, re-paired because the moved files' using blocks now pair differently. One fingerprint (0bb96000…) has count 2. No method-body pair was baselined. The delta is checked against the live clone set: the baseline regenerated from an empty fingerprint set at this head has 811 fingerprints for 823 clones. Every added fingerprint is in it with its exact count, and no removed one is.

Master's other stale entries are left for TST-E0c, which makes stale entries fail the gate (#384). The exception is src.baseline.json's 694a3508…: the moves commit removes it. It is stale on master (this PR touches nothing under src/), and TST-E0c removes the same entry.

Review. A fresh read-only review re-checked:

• the helpers' ownership;
• every moved resource's include and LogicalName;
• the .editorconfig sections;
• stale path references;
• the baseline delta.

Its §3 note and its PathCommandShim defect are fixed in this head. Its Gate finding (Transport-only at d13d9833) no longer holds after #398, so Gate stays at root, as above.

Implementer's run. The implementer's whole-project run showed 38 ServiceFixture failures. They came from a narrow build that skipped packaging, not from a defect; the solution-build run below passes them.

✅ Completion Gate

Local check (quality-gates.md) on head b960774b, base cde7e254, idle host. Every step's own exit code was read:

• slopwatch: zero findings
• dotnet build --configuration Release: 0 warnings, 0 errors
• dotnet format whitespace / dotnet format style --severity warn: no changes
• verify-clones: passed, 0 new
• OpenCode.Sdk.Tests (all but ServiceElection/NativeAot): 8179, 0 failed
• OpenCode.Sdk.Tools.Tests (all but NativeAot): 1794, 0 failed

The host runs Bun 1.4.3, a maintainer-named off-pin host (OPENCODE_SDK_TESTS_BUN_MISMATCH=allow); the pin is 1.4.2.

✅ Code Quality Checklist

• No new suppression (the ledger rows moved with their files)
• No tests skipped, weakened, or deleted to make the suite pass