Agents panel: Separate OpenCode usage into its own collector instead of merging into Claude Code #7747

Summary

The Claude usage collector (omarchy-agent-usage-claude) currently scans OpenCode SQLite history (~/.local/share/opencode/opencode.db) and merges Anthropic-provider messages into Claude Code's local stats and "TOKENS BY MODEL" breakdown.

This causes confusion and data pollution in the Claude Code tab:

1. Model pollution: OpenCode internal, proxy, or experimental model identifiers (e.g. claude-sonnet-4-6, claude-opus-4-6, claude-sonnet-4-5) appear directly in the Claude Code "TOKENS BY MODEL" list.
2. Subscription mismatch: OpenCode sessions often use independent API keys, OpenCode Zen/Go credits, or third-party gateways rather than the Claude Code OAuth subscription. Merging them misrepresents the actual Claude Code usage and limits.

Expected Behavior

• Claude Code collector (omarchy-agent-usage-claude) should only reflect Claude Code CLI native usage (~/.claude/projects/).
• OpenCode usage should be tracked in its own dedicated collector / tab in the Agents panel (similar to the ongoing work in PR Add OpenCode Go collector for the Agents panel #7687 for OpenCode Go) so each tool's usage, models, and quotas remain accurate and isolated.

Steps to Reproduce

1. Use OpenCode with Anthropic provider models (recorded in ~/.local/share/opencode/opencode.db).
2. Use Claude Code separately.
3. Run omarchy-agent-usage-update (or let the agents widget refresh).
4. Open the Agents bar panel on the Claude Code tab.
5. Observe that "TOKENS BY MODEL" lists the model IDs and token counts aggregated from OpenCode.

System details

• Omarchy version: 4.0.0-1
• Kernel: Linux 7.1.8-arch1-3
• Hyprland: 0.56.2
• Component: shell/plugins/agents (omarchy.agents) / bin/omarchy-agent-usage-claude