#6559: Fix LiteLLM reasoning-tag handling + fallback to <think> content

ummary Treat litellm as a reasoning-tag provider. If a model emits only <think>...</think> with no <final>, fall back to the <think> content. Why LiteLLM-proxied models (e.g., Qwen3 via Ollama) often wrap the entire response in <think> without emitting <final>. OpenClaw strips <think> content and returns empty output. This change restores visible replies while preserving existing <final> behavior. Changes provider-utils: add litellm to reasoning tag providers. reasoning-tags: when no <final> exists and stripped output is empty, return the content inside <think> (also handles unclosed <think>). tests added for think-only output. Testing Added unit tests for think-only and unclosed-think inputs. Verified in a local deployment with LiteLLM + Qwen3 (payloads no longer empty). <!-- greptile_comment --> <h2>Greptile Overview</h2> <h3>Greptile Summary</h3> This PR updates reasoning-tag handling so LiteLLM is treated as a “reasoning tag provider” and adjusts `stripReasoningTagsFromText` to fall back to returning `<think>` content when there is no `<final>` and stripping would otherwise yield an empty output. It also adds unit coverage for “think-only” and “unclosed think” responses. The change fits into the existing text sanitation pipeline by ensuring provider-specific tag-wrapped outputs don’t result in empty user-visible responses, while keeping the current behavior of stripping tags when `<final>` is present. <h3>Confidence Score: 2/5</h3> - This PR is risky to merge as-is because it can expose `<think>` (hidden reasoning) content to end users in common inputs. - The new fallback logic returns content from closed `<think>...</think>` blocks whenever there is no `<final>` and the stripped output is empty, which can leak reasoning text rather than only rescuing malformed/unclosed tag cases. The LiteLLM provider detection is also likely too strict if provider strings include prefixes/suffixes. - src/shared/text/reasoning-tags.ts, src/utils/provider-utils.ts <!-- greptile_other_comments_section --> _{(2/5) Greptile learns from your feedback when you react with thumbs up/down!} **Context used:** - Context from `dashboard` - CLAUDE.md ([source](https://app.greptile.com/review/custom-context?memory=fd949e91-5c3a-4ab5-90a1-cbe184fd6ce8)) - Context from `dashboard` - AGENTS.md ([source](https://app.greptile.com/review/custom-context?memory=0d0c8278-ef8e-4d6c-ab21-f5527e322f13)) <!-- /greptile_comment -->