fix(bigtable): test resource leaks, timestamp clamping, and metrics exporter

[bshaffer]

Separated from #9058 to isolate Bigtable CI fixes:

• Fixes resource leaks in Bigtable system tests by preventing reapInstances from deleting active test instances
• Fixes OpenTelemetry timestamp clamping in exporter
• Overrides OpenTelemetry metrics periodic exporter interval to 1s in tests to prevent timeouts

[gemini-code-assist]

Code Review

This pull request updates the client-side metrics exporter to handle cases where the start time is greater than or equal to the end time, and stubs the OpenTelemetry metric reader in tests to use a faster export interval. Additionally, in the system tests, the instance reaping age is increased to 2 hours, and instances are now tagged with a creation timestamp. However, the reviewer correctly points out that changing the reaping logic to ignore instances without a 'time_created' label could lead to resource leaks if other tests create instances without this label.

[westarle]

it's a bit hard to review (e.g. the change in exporter.ts without knowing what the failure mode was.

[bshaffer]

it's a bit hard to review (e.g. the change in exporter.ts without knowing what the failure mode was.

When OpenTelemetry records very fast operations (or during system tests), it can emit metric data points where startTime >= endTime. Google Cloud Monitoring's CreateTimeSeries API was rejecting these with INVALID_ARGUMENT: startTime must be before endTime. So this change just ensures that endTime is at least 1 nanosecond after startTime to prevent those API rejections.