v0.93.0

Every statement is published on a Node diagnostics_channel tracing
channel, so a tracer or APM agent sees Turbine queries with no setup in the
application. No query changes its SQL or its result.

Added

• The turbine:query tracing channel. Each statement that emits a
  $on('query') event is also published on
  diagnostics_channel.tracingChannel('turbine:query'), with
  { model, operation, sql, paramCount, rowCount, durationMs, error? } plus
  batch and retried where they apply. sql is the parametrized text and
  the context never carries a bound value or a result row. The transaction
  control Turbine sends (BEGIN, COMMIT, ROLLBACK, savepoints,
  sessionContext settings) is published too, as model: '$raw',
  operation: '$transaction', and db.$notify as operation: '$notify'.
  start runs in the caller's async context, so the span active there
  becomes the parent. It
  covers model methods (nested with reads included), raw and typed SQL,
  statements inside $transaction, pipelines and batch transactions, and
  works without a $on listener. Without a subscriber a statement costs one
  hasSubscribers check. The name is exported as QUERY_CHANNEL and the
  context type as QueryChannelContext. Published on Node 20.16, 22.3 and
  later; edge runtimes publish nothing and the edge entry imports no new
  builtin.