fix: pass max_retries and timeout_sec from YAML config to LLMClient

[dennis-lynch-nv]

Problem

LlmConfig dataclass and _parse_llm_config() in config.py did not read max_retries or timeout_sec from the YAML config file. Additionally, LLMClient.from_rc_config() in client.py did not forward these values to LLMConfig.

This caused the LLMClient to always use the hardcoded default of 3 retries and 300s timeout regardless of what the user set in config.arc.yaml.

For example, setting max_retries: 8 in config had no effect — the client still retried only 3 times.

Fix

1. config.py — Added max_retries and timeout_sec fields to the LlmConfig dataclass, and parse them in _parse_llm_config()
2. client.py — LLMClient.from_rc_config() now forwards both values to LLMConfig

Verification

from researchclaw.config import load_config
from researchclaw.llm.client import LLMClient

cfg = load_config('config.arc.yaml')  # max_retries: 8, timeout_sec: 1900
client = LLMClient.from_rc_config(cfg)
assert client.config.max_retries == 8   # Previously: 3
assert client.config.timeout_sec == 1900  # Previously: 300

[Jiaaqiliu]

Hi @dennis-lynch-nv, thank you for the contribution! The config passthrough fix is a good idea, but we found two issues in the current diff:

1. Debug print left in production code (line 469): print(f"[DEBUG] LLM call to {url} model={model} max_tokens={max_tokens}") — this would print on every LLM call
2. Dead code: last_err = f"Timeout/OSError: {exc}" after a continue statement (unreachable)

We've merged #234 (your error tracking fix, which is the clean subset of this PR). For the config passthrough (max_retries/timeout_sec), please feel free to resubmit a clean PR with just that change — it's a valid improvement!