Record streaming attempt and front-end metrics when PartialResultSet.last=True is observed #18656

Description

In google-cloud-spanner, streaming queries that receive PartialResultSet.last = True hand off the underlying response iterator to _GLOBAL_STREAM_DRAINER to drain trailing metadata to EOF asynchronously on a background worker thread (in snapshot.py).

However, _StreamingResponseWrapper in metrics_interceptor.py currently defers recording attempt completion (attempt_latencies) and front-end timing metrics (gfe_latencies, afe_latencies) until StopIteration is raised:
https://github.com/googleapis/google-cloud-python/blob/main/packages/google-cloud-spanner/google/cloud/spanner_v1/metrics/metrics_interceptor.py

This causes two issues:

1. Metric Timing Inaccuracy: Attempt latency should measure when the operation actually completes from an outside observer's perspective (i.e. when PartialResultSet.last = True is received by the caller), rather than including time spent waiting in the background drainer queue or consuming trailing stream metadata.
2. Race Condition with Metrics Readers: Callers that complete a query and immediately collect or inspect metrics observe missing attempt_latencies, gfe_latencies, and afe_latencies because the background drainer thread has not yet reached EOF.

Proposed Solution

The client should measure attempt completion and record front-end metrics as soon as PartialResultSet.last == True is observed on the stream, rather than waiting for StopIteration during background stream draining.