feat(web-evals): add multi-model launch and UI improvements

[hannesrudolph]

Summary

This PR updates the evals /runs/new page with several UI improvements and a new multi-model launch feature.

Changes

UI Cleanup

• Removed "Use Multiple Native Tool Calls" checkbox
• Removed "Reasoning Effort" dropdown
• Reorganized sliders into compact rows:
  • 3-column row for Concurrency, Timeout, and Iterations
  • 2-column row for Command Timeout and Shell Integration Timeout

Multi-Model Launch Feature

• Added + button next to model selectors (for Roo Code Cloud/OpenRouter) and config selectors (for Import mode)
• Clicking + adds another model/config selector row
• Each non-last row has a - button to remove that selection
• When multiple models/configs are selected, pressing "Launch" creates identical test runs for each with a 1-minute delay between runs
• After all runs are launched, the user is automatically redirected to the main evals dashboard (/)
• Toast notifications show progress ("Launching X runs (every 20 seconds)..." and "Run X/Y launched" for each)

Testing

• Type checking passes
• ESLint passes

---

Important

Adds multi-model launch feature and UI improvements to evals /runs/new page in new-run.tsx.

• Multi-Model Launch:
  • Added + and - buttons for model/config selectors in NewRun to allow multiple selections.
  • Launches identical test runs for each model/config with a 1-minute delay between runs.
  • Redirects to main evals dashboard after all runs are launched.
• UI Improvements:
  • Removed "Use Multiple Native Tool Calls" checkbox and "Reasoning Effort" dropdown.
  • Reorganized sliders into compact rows: 3-column for Concurrency, Timeout, Iterations; 2-column for Command Timeout, Shell Integration Timeout.
• Misc:
  • Added ModelSelection and ConfigSelection types for handling multiple selections.
  • Updated onSubmit to handle multiple model/config launches with delay and error handling.

^{This description was created by for cd6cbd1. You can customize this summary. It will automatically update as commits are pushed.}