Update MWG description for activation across developer prompts

[micahjo7]

Continued from #456.

We want to activate for relevant web dev tasks- an analysis and experimentation was performed, notes in this doc.

Also maintains 1024-character limit to comply with most coding agents.

This was also tested against our evals to make sure we maintain activation rates (98%+ across three coding agents Claude, Jetski CLI, Codex CLI): https://googlechrome.github.io/guidance/?tests=jetski-cli-mwg-desc-update%7C%7C%7Cremote%2Ccodex-cli-mwg-desc-update%7C%7C%7Cremote%2Cclaude-code-mwg-desc-update%7C%7C%7Cremote

[micahjo7]

Posted another update addressing comments. Also realized there was originally a bug in my testing code, so some other edits were needed. The results are updated in the report:

Total Prompts Evaluated: 98 (with 2 prompts catching JSON parsing escapes)
True Positive (TP) Rate: 0 / 13 (Conversational Research Latency)
True Negative (TN) Rate: 83 / 85 (97.6% Correctly Ignored)
False Positive (FP) / Over-activation Rate: 2 / 98 (2.0% Over-activation)
False Negative (FN) / Under-activation Rate: 13 / 13 (Conversational Research Latency)
GCLI Trigger Boundary Accuracy: 84.7%

and this still has 97%+ activation rate in full suite eval runs across three agents.