Follow-up: SQL dialect and qualified table names in db.query.summary

[Lms24]

Follow-ups for db span names that can only be done once the other PRs in the stack have landed. Tracking them here so they don't get lost.

1. Pass the SQL dialect when sanitizing

#23659 adds a dialect param to sanitizeSqlQuery ('standard' | 'mysql'), because "..." is a string literal in MySQL rather than a quoted identifier, and \ escapes. It updates mysql.ts and mysql2/, but not knex — knex is ported in a different PR and sanitizes before summarizing without passing a dialect, so a knex.raw() statement with an inlined literal on the mysql/mysql2 client can leak a value into db.query.summary.

The dialect is available at the call site: client.driverName is already read there for db.system.name (packages/server-utils/src/integrations/knex.ts). Only mysql/mysql2 map to 'mysql'; pg, sqlite3, mssql and oracledb are all 'standard'.

Worth checking the prisma tracing helper for the same thing — it's multi-connector and also sanitizes with the default.

2. getSqlQuerySummary truncates schema-qualified quoted tables

QUOTED_OR_PLAIN_TABLE_RE in packages/core/src/utils/sql.ts matches the first quoted identifier and stops, while the INSERT/UPDATE/DELETE/DDL branches use TABLE_NAME and keep the whole qualified name:

SELECT ... FROM "public"."User"                 ->  SELECT "public"
DELETE FROM "public"."User"                     ->  DELETE "public"."User"
SELECT ... FROM "public"."A" JOIN "public"."B"  ->  SELECT "public" "public"

Pre-existing — it already affects db.query.summary for postgres/postgresjs — but the db span-name work promotes the value into span names, where the JOIN case makes two different tables indistinguishable.

3. Update the test expectations that encode #2

The prisma v5/v6/v7 node integration tests currently assert SELECT "public" as both the summary and the streamed span name.

Part of #22350. Blocked on #23659 and the rest of the db span name stack landing.

[linear-code]

JS-3450