Backfill should handle rows with missing AnalysisDate gracefully

[kbighorse]

Context

Rows with missing AnalysisDate were previously inserted with a 1970-01-01 epoch sentinel, which is indistinguishable from real data downstream. In #558 these rows are now skipped with an error logged to BackfillResult.errors.

Work Required

Decide on a longer-term strategy for date-less observations:

• Option A: Keep skipping (current behavior) and report to data stewards for manual correction
• Option B: Allow nullable observation_datetime (schema change) for legacy data
• Option C: Derive date from a related field (e.g., sample collection date from NMA_Chemistry_SampleInfo)

Refs

Changed in #558

[github-actions]

Linked Jira ticket created: BDMS-621