Add Qwen inference gateway and current Codex transcript support

[AbsoluteMode]

Session Recall can now use a hosted Qwen inference gateway for both embeddings and reranking. The new inference-api preset sends the document/query distinction, supports compact base64 vectors, validates response dimensions and indices, and keeps the gateway endpoint and credentials configurable. A separate index path and deployment-aware fingerprint allow migration without mixing embedding spaces.

Current Codex desktop sessions record visible messages as completed typed items. The extractor now indexes those user and assistant messages while excluding mirrored messages, tools, and reasoning; its version bump refreshes previously empty sessions.

The setup guide covers credentials, limits, migration, rollback, and MCP restart requirements. Canonical project documentation and synchronized agent guidance were bootstrapped using the documentation workflow. Exact validator allowances cover existing non-secret examples, and a historical personal path was replaced with a portable placeholder.

Validation:

• 478 tests passed; 3 live/model-download tests excluded locally.
• Live embedding and reranking requests, including over-context inputs and base64 vectors, passed against the configured gateway.
• End-to-end MCP search returned semantic results with no degraded fallback.
• Full local migration completed; SQLite integrity and text/vector/FTS row counts matched.
• Documentation validation, guidance synchronization, and diff checks passed.