Add token cost measurement to DOMShell vs CiC benchmark

[apireno]

Goal

Extend the existing DOMShell vs CiC (Claude in Chrome) benchmark with token cost measurement.

Background

The current experiment (experiments/claude_domshell_vs_cic/) measures tool call counts — DOMShell averaged 4.3 API calls per task vs 8.6 for screenshot-based browsing across 4 tasks, 8 trials.

Token cost hasn't been measured yet, but the gap should be even wider than the 2x on call count:

• DOMShell returns structured text (AX node names, roles, properties) — a full tree output for a complex page is ~2-3KB
• Screenshot-based browsing sends base64-encoded page captures — a single screenshot can be 500KB+ in the context window
• Vision tokens are typically priced higher than text tokens

Tasks

• Add input/output token counting to the benchmark harness
• Re-run the existing 4-task suite with token tracking enabled
• Report total tokens per task for both approaches
• Compare cost at current API pricing (text vs vision tokens)
• Update experiments/claude_domshell_vs_cic/ with results

Hypothesis

Structured text (2-3KB per response) vs base64 screenshots (500KB+) should show >2x token savings on top of the 2x call-count reduction already measured.