agent_vectors is unported to Go/TS and unexecuted in Python/C#; cargo test fail-fast masked 36 failures

[sethjuarez]

Summary

28 vectors define agent tool-calling behaviour and no runtime has ever passed them. The emitter-adjacent defect they expose (see companion issue) was reachable the entire time; no backend test matrix was in a position to report it.

runtime	references agent_vectors	status
rust	present	0 passed / 28 failed
go	absent	suite not ported
typescript	absent	suite not ported
python	present	never executed (package does not build)
csharp	present	never executed

Companion: a measurement trap that hid it

Plain cargo test stops after the first failing target, so it only ever reported the lib target. Every "293 passed / 3 failed" figure quoted during the 0.4.19-0.4.21 effort was lib-only and wrong.

Measured on measure/wip-conn (pinned to published @typra/emitter 0.4.21):

cargo test                    ->  293 passed;  3 failed   (lib target only)
cargo test --no-fail-fast     ->  39 failures across 7 targets

Breakdown of the 39:

lib                          3 failed
agent_vectors               28 failed
named_collection_vectors     3 failed   (all: model: missing required field)
+ 4 more single-test targets 5 failed

Any Rust measurement in this repo needs --no-fail-fast. Worth encoding in CI and in the contributing docs.

Requests

1. Port agent_vectors to Go and TypeScript.
2. Get Python and C# executing the vectors they already reference (Python is blocked on the package not building).
3. Add --no-fail-fast to the Rust CI invocation so target-level failures cannot be masked.