[Feature]: Implement streaming content and tool call events

[doubleuuser]

Summary

Fix issues with slow chat responsiveness and less than ideal display of tool call results

Problem statement

I am building a web UI for the ZeroClaw gateway which will have a full workspace, chat, skill and apps etc. I've been working on chat, and implementing different card types for streamingContent, reasoningContent and tool calls.

While doing this I have had two issues:

1. streamingContent card type seemingly not used; agent responses arrive in one large chunk instead of being streamed as it is being generated, which makes the same model feel 10x slower in ZeroClaw compared to VSCode or other clients (same chunking issue observed in terminal)
2. tool call content being displayed content inside regular messages

Here is my audit of current behavior:

Root Causes (Summary)

• RC1: ZeroClaw compat streaming emits one content delta; client must re-chunk for visible progressive rendering.
• RC2: /v1/chat/completions does not convey structured tool telemetry; client needs /api/events bridging to render tool cards for gateway tool use.

F1: “Streaming” can legitimately look non-streaming

Cause:

• ZeroClaw compat POST /v1/chat/completions emits the full assistant text in one SSE delta chunk.
  Effect:
• Client sees exactly one large delta event, so the UI renders the entire answer in one update.
  Conclusion:
• This is not a frontend bug by itself; it is parity with ZeroClaw compat behavior. Client must re-chunk large deltas if it wants visible incremental rendering.

F2: Tool use often cannot render as tool cards when using /v1/chat/completions

Cause:

• ZeroClaw’s /v1/chat/completions compat response does not include structured tool-call events.
• ZeroClaw sanitizes tool-tag markup from assistant text (so the backend cannot reliably recover <tool_call>...</tool_call> from content).
  Effect:
• Even when ZeroClaw uses tools internally, Client may only get plain assistant text describing tool steps, which the UI renders as text.
  Conclusion:
• To render tool-use UI consistently, Client needs a structured tool signal. The only stable source in ZeroClaw is /api/events (SSE broadcast tool telemetry).

Proposed solution

1. Implement streaming of deltas
2. Expose structured tool telemetry

Non-goals / out of scope

N/A

Alternatives considered

No response

Acceptance criteria

• Client can stream agent thinking and reasoning process step by step in web ui
• Client can display tool call progress and results in a separate toolCall component

Architecture impact

/v1/chat/completions

Risk and rollback

Risk: May break existing frontend implementations; however this change is needed to deliver the expected user experience from an agent

Breaking change?

No

Data hygiene checks

• I removed personal/sensitive data from examples, payloads, and logs.
• I used neutral, project-focused wording and placeholders.

[github-actions]

Thanks for opening this issue.

Before maintainers triage it, please confirm:

• Repro steps are complete and run on latest main
• Environment details are included (OS, Rust version, ZeroClaw version)
• Sensitive values are redacted

This helps us keep issue throughput high and response latency low.

[theonlyhennygod]

Closing per maintainer request; implementation tracked in PR #2798 (base dev).