propagate trace processor errors to users instead of internal logging

Remove internal logging of errors in span processors and tracer provider
in favor of returning errors to callers and providing an error handler
callback mechanism for background operations.

Changes:
- Add error_handler callback to BatchSpanProcessor for background errors
  (span drops, export failures during timer-based flushes)
- Return errors from shutdown_with_timeout when spans were dropped
- Remove otel_debug/otel_warn/otel_error calls from span processors
- Simplify export error handling by propagating instead of logging
- Update BatchSpanProcessor builder with with_error_handler() method
- Change dropped_spans_count to Arc<AtomicUsize> for shared ownership
- Remove shutdown error logging from TracerProviderInner

This gives users explicit control over error handling rather than relying
on internal SDK logging that may be missed or not actionable.

Author: svencowart

opentelemetry-sdk/src/trace/provider.rs

@@ -116,14 +116,6 @@ impl TracerProviderInner {
         let mut results = vec![];
         for processor in &self.processors {
             let result = processor.shutdown_with_timeout(timeout);
-            if let Err(err) = &result {
-                // Log at debug level because:
-                //  - The error is also returned to the user for handling (if applicable)
-                //  - Or the error occurs during `TracerProviderInner::Drop` as part of telemetry shutdown,
-                //    which is non-actionable by the user
-                otel_debug!(name: "TracerProvider.Drop.ShutdownError",
-                        error = format!("{err}"));
-            }
             results.push(result);
         }
         results