Analyse transaction durations across feature categories

To understand what is "too slow", we would need to break down the distribution of transaction duration across feature categories, database types and type (sidekiq/webservice).

Possible follow-up to the findings would include

• re-defining/defining application-side SLIs for database transactions
• creating urgency buckets for transactions, similar to sidekiq job urgency
• define a reasonable hard-cap in transaction timeout on the application-side