Fix: preserve double timestamps for PointValueAggregator and unify event start_ts/end_ts type

[tombonfert]

Summary

When a report mixed a PointValueAggregator (PVA) and a StatsAggregator and the channel timestamps were double, the PVA cast point timestamps to long while the StatsAggregator passed double through. This truncated fractional timestamps and, because every event type writes to the shared event_instance_fact table, produced inconsistent column types. In incremental mode this surfaced as a hard failure — DELTA_MERGE_INCOMPATIBLE_DATATYPE: LongType and DoubleType — when a report seeded with long-emitting events (PointsInTimeEvent/ContainerEvent) later had a double-emitting event (BasicEvent) added. (A single full run hid the bug: per-type DataFrames are unioned before the first write, silently widening long→double.)

Changes

• PointValueAggregator — drop the .cast("long") on point timestamps (point_value_aggregator.py).
• PointsInTimeEvent — drop the matching .cast("long") so its event_instance_id hash stays consistent with the PVA (points_in_time_event.py).
• ContainerEvent — cast start_ts/end_ts to double instead of long (container_event.py).

Result: every event type now emits double start_ts/end_ts, so event_instance_fact has a stable double schema from the first write. BasicEvent/SequenceOfEvents already emitted double, so they were unchanged.

Test Plan

• Unit tests added/updated
• Manual testing completed
• Documentation updated (if applicable)

Checklist

• Code follows project style guidelines
• Self-review completed
• No new linter warnings introduced