feat: add summary_only parameter to reduce response size

[ljadach]

Summary

• Adds an optional summary_only boolean parameter (default: False) to 7 query tools
• When enabled, time-series arrays and non-essential metadata are stripped from responses while preserving all key metrics
• Zero overhead on the default path — existing behavior is completely unchanged

Motivation

MCP tool responses contain large time-series arrays (per-minute sleep movement, heart rate readings, HRV data, etc.) that are rarely needed by LLM consumers. For routine health checks, these arrays account for 90%+ of response size while providing no additional value.

For example, query_sleep_data returns ~95K characters for a single night, but the essential metrics (sleep duration, scores, averages) fit in ~2K characters. This wastes context window budget and increases token costs significantly for daily/weekly health monitoring workflows.

Changes

Tool	Estimated Reduction	What's Stripped
query_sleep_data	~95%	Per-minute movement, HR, stress, respiration, HRV arrays
query_health_summary	~65-75%	Duplicate stats call, Body Battery time-series
query_heart_rate_data	~90%	heartRateValues time-series array
query_activities	~60-70%	userRoles, splitSummaries, profile images, connectIQ data
query_weight_data	Minor	Device metadata, source details
get_performance_metrics	Moderate	HRV readings arrays, timestamp details
analyze_training_period	Moderate	Weekly trend breakdown

Implementation

• Blocklist approach for health data (sleep, HR, body battery): strips known time-series keys while passing through any new summary fields Garmin adds
• Allowlist approach for activities and weight: keeps only essential fields, automatically excluding any new verbose fields
• Shared strip_keys() utility in response_builder.py for consistent filtering
• format_activity_summary() in ResponseBuilder for activity-specific field selection
• 6 new unit tests covering strip_keys() and format_activity_summary()

Usage

# Routine daily check — compact response
query_sleep_data(date="today", summary_only=True)

# Ad-hoc investigation — full detail (default, unchanged behavior)
query_sleep_data(date="today")

Test plan

• All 64 existing + new unit tests pass
• ruff check and ruff format pass
• Default behavior (summary_only=False) is identical to pre-change behavior
• Manual testing with live Garmin data to verify summary responses contain all needed metrics

🤖 Generated with Claude Code