add native DeepSeek support and robust binary deployment

[lee101]

What changed

• add a built-in DeepSeek provider for deepseek-v4-flash and deepseek-v4-pro
• implement native OpenAI-compatible Chat Completions requests, streaming reasoning, and tool-call handling
• infer the DeepSeek provider from either supported model name and expose native model metadata
• make packaged binary deployment verify and include codex-code-mode-host

Why

DeepSeek's API uses Chat Completions rather than the Responses transport used by Codex's default provider. The local package also allowed the code-mode host to be absent, causing runtime failures after rebuilding only the main binary.

Impact

Users can export DEEPSEEK_API_KEY and run either DeepSeek V4 model directly. Release builds now package and validate the code-mode companion host.

Validation

• scoped Rust test suites and cargo check -p codex-core
• clippy fixes for all affected crates
• generated config schema and thread-config protobuf checked
• optimized release build for codex-cli and codex-code-mode-host
• vendored Linux x64 package verification
• live deepseek-v4-flash request returned the expected response

[codex-infinite]

Codex Infinity
Hi! I'm Codex Infinity, your coding agent for this repo.

Start a task on this PR's branch by commenting:

• @codeggsinfinity review
• @codeggsinfinity fix the failing tests
• /codex

Tasks and logs: https://codex-infinity.com