Clean and Normalize Data Streams

Amgen's laboratory systems can emit sample events with inconsistent field names, capitalization, whitespace, timestamp formats, and numeric units. Implement normalize_stream to clean a batch of incoming event records and return only valid, canonical events.

Formal Specification

Input is a list of dictionaries. Each record may use these aliases:

• Event ID: event_id or id
• Sample ID: sample_id or sample or specimen_id
• Status: status or state
• Dose: dose_mg or dose
• Timestamp: timestamp or time

Return a list of dictionaries with exactly these keys: event_id, sample_id, status, dose_mg, and timestamp.

Normalize values as follows:

1. Trim identifier whitespace and convert identifiers to strings.
2. Convert statuses to lowercase and map complete and completed to completed, in-progress and processing to in_progress, and failed and error to failed.
3. Parse doses such as "12.5 mg", "1,000", or numeric values into nonnegative floats. Missing doses become None.
4. Parse ISO-8601 timestamps. Treat timestamps without a timezone as UTC and return UTC timestamps ending in Z.
5. Discard records missing required fields, having unknown statuses, invalid timestamps, or invalid doses.
6. If multiple valid records share an event_id, keep the last valid record while preserving the order of each event's first appearance.