fix(dashboard-api): recognize reasoning tokens and tool calls in pixel chat stream

[vaibhavsrv]

Why this matters

In ods/extensions/services/dashboard-api/routers/pixel.py, the retained chat stream handler _produce_retained_result() monitors Server-Sent Events (SSE) from the Pixel edge service to determine whether a response contains meaningful output before saving it into ChatResultStore. Lines 780-783 evaluate answer_seen by strictly checking if isinstance(payload, dict) and isinstance(payload.get("content"), str) and payload["content"]: answer_seen = True. When using reasoning models (such as DeepSeek-R1 distillations, Qwen reasoning checkpoints, or models emitting thought traces) or tool-calling models, generated tokens are streamed in reasoning_content (or tool_calls) while content remains empty or omitted. Because answer_seen remains False, the arrival of data: [DONE] (lines 784-794) misclassifies the completed response as an aborted/zero-output session, flags failed = True, and overwrites the user transcript with an error frame: "Pixel returned no answer. Try again.".

This fix surgically expands the stream content probe in _produce_retained_result() to set answer_seen = True when payload contains non-empty reasoning_content or tool_calls. Genuine zero-token aborts (such as premature cancellations or role-only frames) remain rejected as empty, while reasoning streams and tool executions complete successfully. Existing receipts persistence, ownership boundaries, and stream timeouts remain untouched.

Validation

• Baseline reproduction: Upstream SSE streams emitting reasoning_content chunks followed by data: [DONE] evaluated answer_seen as False, marking the session state as interrupted and appending "Pixel returned no answer. Try again.".
• Post-fix behavior: Upstream streams emitting reasoning_content or tool_calls resolve cleanly to state complete without injecting false errors, while role-only streams without tokens continue to be flagged as interrupted.
• Telemetry statement: "[Dashboard-api] suites: 3 passed. New-test Ruff, ShellCheck, and diff checks pass; new regressions wired into Linux CI."

Overlap check

• PR fix(model-contract): coerce integer float token limits in projection #5606 (fix(model-contract): coerce integer float token limits in projection): Addressed integer float coercion in ods/bin/pixel_model_contract.py; does not touch dashboard SSE stream consumption.
• PR fix(pixel-relay): return 502 bad gateway when upstream model router is unavailable #5605 (fix(pixel-relay): return 502 bad gateway when upstream model router is unavailable): Addressed upstream connectivity error boundaries in pixel-model-relay; does not touch dashboard receipt tracking.
• PR fix(pixel-edge): accumulate content before finish reason in stream fallback guard #5604 (fix(pixel-edge): accumulate content before finish reason in stream fallback guard): Focused on edge runtime buffer accumulation; does not alter dashboard-api retained stream event parsing.

Risk / AI disclosure

AI-assisted investigation, implementation and CLI regressions. This strengthens SSE stream payload accounting for reasoning models, not runtime admission. Independent human review and platform/runtime qualification remain gates. No running configuration, deployment or upstream merge changed.

Follow-up integration evidence

Composed with #5606 at ebe8577 without conflicts. Production and test diffs passed together; pixel chat results and stream checks remain intact.
Backlog composition was local-only (production/test diffs, excluding workflow/Makefile wiring); it is not an upstream merge or independent human approval. Declared live-review gates remain open.