fix(agent): avoid full availability refresh on reads

[kaizhou-lab]

Summary

• Remove the legacy full-agent availability refresh path from normal reads, startup hydration, custom-agent mutations, and the unused /api/agents/refresh endpoint.
• Keep agent availability state cache-backed: manual health checks only update the requested agent, and agents without a probe result are surfaced as Unchecked instead of being assumed available.
• Add API status support for Unchecked so the frontend can display unknown/unprobed availability accurately.
• Fix backend daily log rotation so cross-day writes go to the matching YYYY/MM/DD/YYYY-MM-DD.{aioncore,aionrs}.log directory instead of writing the new date under the previous day directory.

Why

Opening or refreshing the Agents page was triggering repeated full availability probes for all agents. That produced thousands of repeated node runtime probe decided / agent available / agent unavailable log lines and unnecessary path/runtime checks. After removing the full refresh, unchecked historical rows need to remain visibly unknown rather than silently becoming available.

A separate log-rotation issue also made cross-day debugging confusing: files such as 2026-07-03.aioncore.log could appear under 2026/07/02/.

Notable Changes

• Management and assistant list reads now reuse stored availability snapshots instead of probing all built-in agents.
• Explicit single-agent health checks no longer invalidate and rehydrate the entire registry.
• The dead /api/agents/refresh endpoint was removed.
• Availability DTOs now include Unchecked for rows that have not been probed.
• Daily log writers compute the local date on each write and reopen the file when the day changes.

Verification

• cargo fmt --all -- --check
• cargo test -p aionui-app bootstrap::tracing_init::tests
• cargo test -p aionui-ai-agent
• cargo clippy -p aionui-app -p aionui-ai-agent -- -D warnings