Problem

The current telemetry implementation has two issues:

1. Global logfire configuration: logfire.configure() sets up the global OpenTelemetry provider, potentially overwriting the user's existing configuration
2. Global request instrumentation: logfire.instrument_requests() monkey-patches all requests.Session usage application-wide, not just garth's

This means garth's telemetry conflicts with applications that have their own logfire/OpenTelemetry setup.

Proposed Solution

Use requests.Session hooks instead of global instrumentation:

1. Attach a response hook directly to garth's session instance
2. Send telemetry data to logfire cloud via HTTP API (not via OpenTelemetry SDK)
3. Keep the same environment variables as defaults:
   • GARTH_TELEMETRY=true/false
   • LOGFIRE_TOKEN
   • GARTH_TELEMETRY_SERVICE_NAME

Benefits

• Isolation: Only garth's requests are logged, no global side effects
• No conflicts: Works alongside existing logfire/OpenTelemetry setups
• Same defaults: Logfire cloud remains the default destination with same env vars
• Customizable: Users can provide their own callback for custom destinations

API Design

# Default: sends to logfire cloud (same as today)
garth.configure(telemetry_enabled=True)

# Custom callback for alternative destinations
garth.configure(
    telemetry_enabled=True,
    telemetry_callback=lambda data: my_logger.info("garth", **data)
)

Implementation Notes

• Remove logfire.configure() and logfire.instrument_requests() calls
• Add session.hooks['response'].append(...) in Client.__init__ or configure()
• Default callback sends sanitized data to logfire cloud via their HTTP ingest API
• Keep all existing sanitization logic