chore(evals): Rename workflow evals to MCP agent evals (#1312)

closes #1286 

## What

Renames the harness to say what it evaluates: an agent using the Apify
MCP server.

- `evals/workflows/` → `evals/mcp_agent/`
- `run_workflow_evals.ts` → `run_mcp_agent_evals.ts`,
`workflow_judge.ts` → `mcp_agent_judge.ts`
- `evals:workflow` → `evals:mcp-agent`, `evals:workflow:export-dataset`
→ `evals:mcp-agent:export-dataset`. No aliases kept.
- Symbols and prose: `WORKFLOW_DATASET_NAME` → `MCP_AGENT_DATASET_NAME`,
`parseWorkflowItem` / `toWorkflowTestCase` / `fetchWorkflowCases` →
`…McpAgent…`, `WorkflowItem` / `WorkflowCase` / `WorkflowTaskOutput` /
`WorkflowTaskOptions` → `McpAgent*`, README and help text.
- Langfuse dataset `workflow-evals` → `mcp-agent-evals` (renamed in the
UI; the public API has no rename endpoint).

Pure rename: `dataset_snapshot.json` moves path with zero content
change.

## Beyond the issue's list

The judge score `workflow_judge` is renamed to `mcp_agent_judge`.

Left alone as out of scope: the Langfuse project name `MCP Workflow`,
the `workflow-evals-rubric` dataset, and the case ids
`workflow-mcp-discover-and-execute` /
`workflow-search-fetch-schema-call`.

## Verification

`pnpm run type-check`, `pnpm run lint`, `pnpm run test:unit` (1455
passed), `pnpm run format`, `pnpm run check:agents` all pass.

## AI assistance

Written with Claude Code: file renames and the symbol and documentation
rewrites. Reviewed by me.