fix(llm): preserve native provider options

[rekram1-node]

Summary

• preserve OpenAI-compatible assistant reasoning continuation fields through native request lowering
• align native OpenAI-compatible, Responses, and Anthropic option lowering with AI SDK behavior used by opencode
• cover DeepSeek tool continuations, Responses encrypted reasoning options, and Anthropic adaptive thinking behavior

Problem

A native OpenAI-compatible tool continuation could omit prior assistant reasoning_content, causing DeepSeek thinking-mode requests to fail with HTTP 400. Auditing the local AI SDK source also found additional native option gaps for Responses and Anthropic models that opencode currently emits.

Changes

• expose transformed compatible message metadata in the shape consumed by native Chat lowering
• support reasoning_content and reasoning_details continuations and streaming deltas
• translate documented compatible options (reasoningEffort, textVerbosity) while passing provider-specific compatible extensions through
• preserve OpenAI Responses include, cache, reasoning, service, and continuation-related wire options
• support Anthropic adaptive thinking, effort output configuration, and enabled-thinking token/parameter rules
• update the native Zen recording to include the previously dropped encrypted reasoning request field

Verification

• cd packages/llm && bun test test/provider/openai-compatible-chat.test.ts test/provider/openai-responses.test.ts test/provider/openai-chat.test.ts test/provider/anthropic-messages.test.ts && bun typecheck
• cd packages/opencode && bun test test/session/llm-native.test.ts test/session/llm.test.ts test/session/llm-native-recorded.test.ts && bun typecheck
• git diff --check

[lc8882972]

尽快合并这个分支吧😭

[rekram1-node]

@lc8882972 just do this:

OPENCODE_EXPERIMENTAL_NATIVE_LLM=0

If u have that env var set this code path isnt ever hit (it isnt hit by default but this will ensure its not hit otherwise either).