chore: remove legacy flat results.jsonl support and align naming around dataset/testcase

[christso]

Summary

AgentV should finish the transition to canonical run workspaces and consistent terminology.

Current active design uses:

• .agentv/results/runs/<run-id>/index.jsonl as the canonical persisted run layout
• dataset as the persisted result field and intended filter term
• test case as the more accurate concept name for individual cases

Follow-up cleanup should remove remaining support for legacy flat results.jsonl inputs and align mixed internal naming that still uses terms like suite, evalSetName, and evalCase.

Why

This reduces design drift and avoids carrying legacy compatibility into new commands like agentv trend.

Scope

Remove legacy result layout support

Audit CLI/result-loading surfaces and remove support for legacy flat result file inputs where we still accept them.

Target canonical input shape:

.agentv/results/runs/<run-id>/index.jsonl

Align terminology

Proposed renames:

Current	Proposed	Why
suite	dataset	aligns with persisted result field and CLI filter term
evalSetName	datasetName	same reason
evalCase	testCase	object is a test case, not an eval run
evalCases	testCases	same reason
rawTestcases	rawTestCases	same reason plus consistent casing
legacy evalId fallback naming	testId only where possible	reduce dual-term confusion

Non-Goals

• Do not change external wire format away from dataset/test_id
• Do not block feature work like #913
• Do not bundle risky behavioral changes unrelated to result input layout or naming cleanup

Acceptance Signals

• remaining legacy flat result-file compatibility is removed or explicitly isolated behind a deliberate compatibility boundary
• new commands only document and accept canonical run workspace inputs
• variable naming in touched areas is aligned toward dataset and testCase
• docs and code comments use consistent terminology