fix(qe): fix error logging

[miguelff]

Fixes #16226

This patch ensures that errors resulting in a failed query generate an event if the error logger is enabled.

This aligns with the principle of least astonishment: The user expects errors to be logged, and due to the lack of a formal definition of what constitutes an error, the most sensible approach in my opinion (backed by bug reports like #14933) is to log any condition leading to an error in the processing of the query, which is, in turn, any of the four well-known classes of errors.

This is done now in the RequestHandler thus capturing any possible error condition happening during query processing. Before, every different engine was doing this differently, and errors were not properly captured.

A smoke test suite proves that NotFoundError errors in findXOrThrow are now properly emitted. I consider this enough, given that the code that handles the error is the same for every exception thrown during query execution.

[miguelff]

Did the following:

• Tampered the binary engine:

diff --git a/query-engine/request-handlers/src/graphql/handler.rs b/query-engine/request-handlers/src/graphql/handler.rs
index 39cb2e08e1..15690c5b47 100644
--- a/query-engine/request-handlers/src/graphql/handler.rs
+++ b/query-engine/request-handlers/src/graphql/handler.rs
@@ -26,7 +26,6 @@ impl<'a> GraphQlHandler<'a> {
 
     pub async fn handle(&self, body: GraphQlBody, tx_id: Option<TxId>, trace_id: Option<String>) -> PrismaResponse {
         tracing::debug!("Incoming GraphQL query: {:?}", &body);
-
         match body.into_doc() {
             Ok(QueryDocument::Single(query)) => self.handle_single(query, tx_id, trace_id).await,
             Ok(QueryDocument::Multi(batch)) => match batch.compact(self.query_schema) {
@@ -44,7 +43,7 @@ impl<'a> GraphQlHandler<'a> {
 
     async fn handle_single(&self, query: Operation, tx_id: Option<TxId>, trace_id: Option<String>) -> PrismaResponse {
         use user_facing_errors::Error;
-
+        dbg!("Running single {}", query.clone());
         let gql_response = match AssertUnwindSafe(self.handle_graphql(query, tx_id, trace_id))
             .catch_unwind()
             .await
@@ -201,6 +200,9 @@ impl<'a> GraphQlHandler<'a> {
         tx_id: Option<TxId>,
         trace_id: Option<String>,
     ) -> query_core::Result<ResponseData> {
+        if 2 > 1 {
+            panic!("Forced")
+        }
         self.executor
             .execute(tx_id, query_doc, self.query_schema.clone(), trace_id)
             .await

• Ran Prisma against the binary engine. The connection is reset before the actual panic is processed asynchronously. In this case, the error is correctly received by the client

• Ran Prisma against the library engine. The behavior is correct.

I think this is ready for another pass.

[aqrln]

One comment/question, but other than that, looks very good to me! Thank you for the fix and for the nice refactoring!