Populate OpenCode skills and subagents breakdowns

[KevNev19]

What

Populate the Skills & Agents breakdown for OpenCode sessions. Before this change the panel always showed "No skill/agent usage" for OpenCode, even when skills and subagents were used — while the Core Tools panel correctly counted the Skill and Agent tool invocations.

Fixes #556.

Root cause

buildAssistantCall (in src/providers/session-message.ts, shared by the OpenCode SQLite parser, the OpenCode file-based parser, and Kilo Code) built the tools list from tool parts but never read the name of the skill / subagent from the tool-call input. The resulting ParsedProviderCall carried no skills and no subagentTypes, so:

• providerCallToTurn / providerCallToCachedCall hard-coded skills: [], so the classifier never set subCategory → empty skillBreakdown.
• subagentTypes was undefined → empty subagentBreakdown.

In OpenCode's on-disk format the identifier lives in state.input:

// part file, skill tool
{ "type": "tool", "tool": "skill", "state": { "input": { "name": "commit" } } }
// part file, task (subagent) tool
{ "type": "tool", "tool": "task", "state": { "input": { "subagent_type": "explore", "description": "..." } } }

Change

• ParsedProviderCall gains an optional skills?: string[] (mirrors the existing subagentTypes?).
• buildAssistantCall extracts skills (from skill tool input.name) and subagentTypes (from task tool input.subagent_type) using .filter/.map (no bracket-assign, so the hot-path Semgrep rule is satisfied).
• providerCallToTurn and providerCallToCachedCall pass call.skills ?? [] instead of [].

The downstream classifier subCategory gate (category === 'general') is intentionally left unchanged — it already has dedicated test coverage and is shared with the Claude path. This PR only fills the missing upstream extraction; subagents (no gate) always show, and skills show for general-category turns exactly as they do for Claude.

Tests

• New fixture tests in tests/providers/opencode-file.test.ts: skill/subagent extraction from skill/task parts, and the empty case.
• Full suite green: 1302 tests, 106 files.
• npx tsc --noEmit clean; npm run build succeeds.

Verification on real data

Ran the built CLI against a real OpenCode file-store. Previously empty, now populated:

SKILLS:    pipeline-investigation, plan-spec, teams, mermaid, newrelic, splunk, jira, tdp-pipelines-report, confluence
SUBAGENTS: explore (189), general (14), general-purpose (2)