Add observability: metrics, tracing, and DLQ support #6

Problem

The Lambda function has no observability beyond slog.Info/slog.Error log lines:

1. No metrics — There is no way to track:

   • Number of records processed per invocation
   • Number of records filtered/dropped
   • Time spent in each pipeline stage (S3 fetch, KMS decrypt, decompress, Parquet convert, S3 write)
   • Error rates by stage

2. No structured error aggregation — When a batch has multiple failures, each error is logged individually but there is no summary of batch-level outcomes.

3. No Dead-Letter Queue (DLQ) handling — If SQS retries are exhausted, messages go to the DLQ silently with no alerting or reprocessing path.

4. No tracing — No AWS X-Ray or OpenTelemetry integration to trace the full pipeline across S3 → KMS → Parquet → S3.

Proposed Solution

Metrics (using embedded metric format via aws-lambda-go or CloudWatch EMF)

Add structured metrics for:

• RecordsProcessed (count)
• RecordsFiltered (count)
• RecordsDropped (count)
• ProcessingDuration (per stage, in milliseconds)
• Errors (count, by stage)

Tracing

• Add aws-lambda-go X-Ray tracing wrapper
• Trace segments for: S3 GetObject, KMS Decrypt, ESDK Decrypt, Decompress, Parquet Convert, S3 PutObject

DLQ

• Document DLQ configuration in README
• Add a reprocessing handler or document manual reprocessing steps

Acceptance Criteria

• CloudWatch metrics are emitted for record counts and processing durations
• X-Ray tracing wraps the handler with sub-segments per pipeline stage
• DLQ setup is documented
• Metrics are testable (not hardcoded to CloudWatch SDK)