feat: automatically enable non-interactive mode for AI agents

[joehan]

This PR automatically sets options.nonInteractive = true when an AI agent is detected in the environment. This makes the CLI more usable for agents by failing fast instead of hanging on prompts.

Evaluation Results

We ran the complete 51-case evaluation suite in google3 (firebase_tools_evals) to compare the baseline (origin/main) against this branch:

Metric	Baseline (main)	Branch (enable-noninteractive)	Change
Pass Rate	51.0% (26/51)	64.7% (33/51)	+13.7% (+7 cases)
Input Tokens	7,489.8k	10,413.4k	+39.0%
Output Tokens	259.8k	420.6k	+61.9%
Total Tokens	7,749.6k	10,834.0k	+39.8%
Total Turns	1,553	2,197	+41.5%
Tool Calls	684	838	+22.5%

Note on Token Usage

Automatically enabling non-interactive mode increases token usage and turns because it prevents the agent from hanging on prompts. Instead of waiting and timing out, the agent receives immediate feedback and can use its remaining turns to actively troubleshoot and find alternative non-interactive flags (e.g. appending --project or --non-interactive explicitly). This active recovery is what drove the 13.7% increase in pass rate.

[gemini-code-assist]

Code Review

This pull request updates src/command.ts to import and use the detectAIAgent utility from ./env. It modifies the condition for setting non-interactive mode to automatically enable it if an AI agent is detected (i.e., detectAIAgent() !== 'unknown'). There are no review comments, so I have no feedback to provide.