WIP: feat: metrics

[jpnurmi]

Summary

Implements the Sentry Metrics feature for the native SDK, allowing applications to record counter, gauge, and distribution metrics that are batched and sent to Sentry: https://develop.sentry.dev/sdk/telemetry/metrics/.

Closes #1453

Features

• Three metric types: counter, gauge, and distribution
• Batching: Metrics are buffered and sent in batches (same logic as logs - every 100 items or 5 seconds)
• Attributes: Automatic default attributes (environment, release, SDK info) plus custom user attributes
• Trace context: Metrics are automatically associated with the current trace/span
• Hook: before_send_metric callback for filtering or modifying metrics before sending
• Unit constants: Pre-defined SENTRY_UNIT_* constants for common telemetry units (byte, kilobyte, millisecond, second, etc.)
• Parallel flush/close: Logs and metrics flush and shutdown operations run in parallel for faster sentry_flush() and sentry_close()

Public API

// Enable metrics
sentry_options_set_enable_metrics(options, true);

// Record metrics
sentry_metrics_count("api.requests", 1, sentry_value_new_null());
sentry_metrics_gauge("memory.usage", 1024, SENTRY_UNIT_BYTE, sentry_value_new_null());
sentry_metrics_distribution("response.time", 150.5, SENTRY_UNIT_MILLISECOND, attributes);

Considerations

The current API requires passing sentry_value_new_null() when no custom attributes are needed. An alternative would be to provide simplified variants without the attributes parameter (e.g., sentry_metrics_count()) alongside explicit variants (e.g., sentry_metrics_count_with_attributes()). Feedback welcome on which approach is preferred.

Dependencies

• ref(logs): extract reusable batcher for metrics #1495
• ref(logs): extract attribute helpers for metrics #1497

Test plan

• Unit tests for all metric types (tests/unit/test_metrics.c)
• Integration tests for HTTP transport (tests/test_integration_http.py)
• Tests for custom attributes, before_send_metric hook, and disabled state
• Crash-safe flush behavior tested

🤖 Generated with Claude Code