feat: improve watermark calculation by adding the option to cache it

[AdityaHegde]

This PR centralises the calculation of min, max and watermark to executors.

• Adds Executor.Timestamps that does the calculation of min, max (previously in metrics_time_range resolver) and watermark.
• Updates metrics_time_range resolver to call Executor.Timestamps.
• Adds Executor.BindQuery that allows for setting the timestamps from outside. This allows for getting a cached timestamps by getting it from metrics_time_range resolver.
• Update MetricsViewTimeRange to use metrics_time_range resolver.
• Update usages of Executor.Query to set cached timestamps.

A future PR will refactor metrics_sql to move time range resolution to Executor. So it is not updated to use the new resolver.

[begelundmuller]

This ends up sending **time.Time to rows.Scan, might not be well handled by all implementations. Also makes the pointers unsafe to access since they may still be nil. I think this would be safer:

Suggested change

	var minTime, maxTime, watermark *time.Time
	err = rows.Scan(&minTime, &maxTime, &watermark)
	var minTime, maxTime, watermark time.Time
	err = rows.Scan(&minTime, &maxTime, &watermark)

[AdityaHegde]

This is needed for duckdb. It returns null when there are no rows, so we need minTime, maxTime, watermark to be pointers. Perhaps that is why we had different implementation for duckdb. We should probably go back to that. What do you think?

[begelundmuller]

Have you tried using WeakDecode? Then I think it would be able to directly decode ISO strings into time.Time types (but I may be wrong)

[AdityaHegde]

It wont work by itself. We need to create a decoder with DecoderConfig and StringToTimeHookFunc. I just reused the code from metricsview_aggregation since we will also need for rilltime resolver.