fix(skill): drop parallel_tool_calls=False from tool-less eval agents

The eval-set generator, trigger grader, and coverage grader have no
tools — they only produce text. OpenAI's API rejects
`parallel_tool_calls` when `tools` is unset:

  Invalid value for 'parallel_tool_calls': 'parallel_tool_calls' is
  only allowed when 'tools' are specified.

`skill eval` would die at the very first LLM call, before any prompt
was graded. Tests masked it because they patch `Runner.run` and never
hit the LiteLLM layer.

Fix: stop passing `model_settings=ModelSettings(parallel_tool_calls=False)`
on those three agents — the flag is both invalid and meaningless when
the agent has zero tools to parallelise. `ModelSettings` is no longer
referenced anywhere in this module, so the import is dropped too.

Pre-existing bug carried over from the original evaluator design; this
is the first time anyone has run `skill eval` end-to-end against a
real LLM (everything else mocks Runner.run).

Author: KylinMountain

openkb/skill/evaluator.py

@@ -39,7 +39,6 @@
 
 from agents import Agent, Runner
 from agents.exceptions import MaxTurnsExceeded
-from agents.model_settings import ModelSettings
 
 from openkb.skill import extract_body, extract_frontmatter
 
@@ -207,7 +206,6 @@ async def generate_eval_set(
         name="eval-set-generator",
         instructions=instructions,
         model=f"litellm/{model}",
-        model_settings=ModelSettings(parallel_tool_calls=False),
     )
     try:
         result = await Runner.run(agent, "Generate the eval set now.", max_turns=3)
@@ -266,7 +264,6 @@ async def grade_one(
         name="trigger-grader",
         instructions=instructions,
         model=f"litellm/{model}",
-        model_settings=ModelSettings(parallel_tool_calls=False),
     )
     try:
         result = await Runner.run(agent, f"Question: {question}", max_turns=2)
@@ -315,7 +312,6 @@ async def grade_coverage(
         name="coverage-grader",
         instructions=instructions,
         model=f"litellm/{model}",
-        model_settings=ModelSettings(parallel_tool_calls=False),
     )
     try:
         result = await Runner.run(agent, f"Question: {question}", max_turns=2)