Update OpenAI default model to gpt-5.4-mini

[timrogers]

Motivation

• Move the OpenAI-mode default to a more modern model string while keeping existing behaviour and provider selection logic unchanged.
• Ensure README and test fixtures reflect the chosen default so documentation and mocked responses stay consistent.
• Keep the change narrow (string-level model update) to avoid API-surface or runtime changes.

Description

• Change the OpenAI default model constant from gpt-4o to gpt-5.4-mini in src/llm.rs (OPENAI_MODEL).
• Update the README to document that OpenAI-mode inference uses gpt-5.4-mini by default for receipt and category inference.
• Refresh LLM test fixtures and inline mocked /chat/completions responses in tests/fixtures/llm_receipt_inference_response.json, tests/fixtures/llm_category_inference_response.json, and tests/llm_api.rs to reference gpt-5.4-mini.

Testing

• Ran cargo fmt --all, which completed successfully.
• Ran git diff --check and basic repo checks (no formatting or diff errors reported) and observed rustc --version in the environment.
• Attempts to run cargo test --locked --all-features and cargo build --release --locked were blocked because the local rustc is 1.89.0 while the crate and dependencies require MSRV 1.94, so full test/build could not be executed here; pre-commit run --all-files was not run because pre-commit is not installed in the environment.

---

Codex Task