Avoid redundant creation of IntervalResults objects in result conversion logic

[bharathappali]

In convertUpdateResultsAPIObjToExperimentResultData(), a new IntervalResults object is being created inside the loop for each metric. However, only the last created object is retained in the resultsMap, making all previously created IntervalResults objects redundant and subject to garbage collection.

This leads to unnecessary object creation and temporary memory pressure.

Link for code

Proposed Fix:

Refactor the logic to:
- Create the IntervalResults object once per container (per interval).
- Populate it with the metricResultsMap after collecting all metrics.
- Then store it in the resultsMap.

This reduces GC overhead and improves performance.

Memory breakdown of IntervalResults object:

Field	Type	Size (Compressed)	Size (Uncompressed)	Notes
Object Header	JVM internal	12 bytes	16 bytes	Includes mark word and class pointer
metricResultsMap	HashMap<MetricName, MetricResults>	4 bytes	8 bytes	Reference only
acceleratorMetricResultHashMap	HashMap<MetricName, AcceleratorResult>	4 bytes	8 bytes	Reference only
intervalStartTime	Timestamp	4 bytes	8 bytes	Reference only
intervalEndTime	Timestamp	4 bytes	8 bytes	Reference only
durationInMinutes	Double	8 bytes	8 bytes	Primitive (boxed if used externally)
duration_in_seconds	Double	8 bytes	8 bytes	Primitive
Total (without padding/alignment)	—	44 bytes	64 bytes	Minimum memory before alignment/padding

Actual heap allocation will be rounded up to 8 or 16 byte alignment, so practical sizes may be 48 bytes (compressed) and 64 or 72 bytes (uncompressed).

So the total temporary memory saving would be 48 bytes (As compressed refs are turned on by default in JVM) per request / per timestamp / per kubernetes object / per container / per metric