Count otel.sdk.processor.{span,log}.processed at exporter-submit time

[cijothomas]

Aligns Python with the processor processed boundary clarified in open-telemetry/semantic-conventions#3902.

The otel.sdk.processor.span.processed / otel.sdk.processor.log.processed counters are currently incremented after exporter.export() returns, and an exporter failure is stamped onto this metric as error.type (the raised exception's class name). That conflates exporter outcome with processor outcome and produces unbounded error.type cardinality.

This aligns Python with the boundary clarified in semantic-conventions#3902: the processor counts a record as processed when it submits the batch to the exporter, regardless of the export result. error.type on this metric is reserved for the processor's own drop reasons (queue_full, already_shutdown); exporter success/failure is reported separately by the otel.sdk.exporter.* metrics.

Changes:

• finish_items(count) no longer takes an error and always records success.
• Batch and simple span/log processors count at submit-time, before export.
• Simple log processor no longer counts a record dropped by the shutdown early-return as processed.