feat: add generic /v1/completions client and token debug visualization (#428)

* feat: add generic /v1/completions client and token debug visualization

Add a generic FireworksV1CompletionsClient that handles local tokenization
via HuggingFace transformers and calls the /v1/completions endpoint with
token-in/token-out. Tool-call parsing is pluggable via a callback rather
than hardcoded to any specific domain.

Also add TokenDebugView component for the evaluation dashboard with:
- Text view: readable colored text with mask overlay (prompt vs completion)
- Episode view: per-token visualization with mask or logprob coloring
- Turns view: per-turn breakdown of prompt/completion tokens
- Token ID chips with hover tooltips showing detokenized text and logprob
- Smooth gradient logprob coloring (green=high confidence, red=low)

Made-with: Cursor

* fix: address CI failures and review comments

- Rewrite tests to match the generic client API (remove references to
  old domain-specific methods like _parse_tool_call_with_optional_fallback)
- Fix TokenDebugSection guard to also check extra?.full_episode
- Fix zero reward styled as red/negative — now uses neutral gray
- Fix tools=[] vs None: explicit empty list no longer falls back to
  default_tools

Made-with: Cursor

* fix: pass thinking kwargs in build_assistant_turn_token_ids, handle missing fullEpisode

- build_assistant_turn_token_ids now passes _thinking_kwargs() to
  apply_chat_template, consistent with _build_prompt_token_ids and
  build_tool_response_suffix_token_ids
- View selector no longer silently renders nothing when fullEpisode is
  null and user selected text/episode view — falls back to turns view
  or shows a placeholder message

Made-with: Cursor

* fix: guard against None after Mapping extraction, prevent request_params overriding core fields

- Add second None check in _normalize_token_id_sequence after extracting
  from a Mapping, since .get() can return None even when the key exists
- Move request_params spread before explicit keys (model, prompt,
  temperature, max_tokens, logprobs) so they cannot be silently
  overridden by stray entries in request_params

Made-with: Cursor

Author: benjibc

eval_protocol/integrations/__init__.py

@@ -3,5 +3,17 @@
 from .openeval import adapt
 from .trl import create_trl_adapter
 from .openai_rft import build_python_grader_from_evaluation_test
+from .fireworks_v1_completions_client import (
+    FireworksV1CompletionsClient,
+    ParsedToolCall,
+    to_openai_tool_calls,
+)
 
-__all__ = ["adapt", "create_trl_adapter", "build_python_grader_from_evaluation_test"]
+__all__ = [
+    "adapt",
+    "create_trl_adapter",
+    "build_python_grader_from_evaluation_test",
+    "FireworksV1CompletionsClient",
+    "ParsedToolCall",
+    "to_openai_tool_calls",
+]