perf(database): reuse prepared statements on the inference hot path

.config/nextest.toml

@@ -32,7 +32,7 @@ max-threads = 4
 # These tests intentionally touch database-wide state (DDL, broad encryption
 # scans, or ACCESS EXCLUSIVE locks), so they must not overlap any other test.
 [[profile.default.overrides]]
-filter = 'package(api) & binary(e2e_all) & (test(/^database_encryption::/) | test(/^reporting_usage::usage_export::usage_export_database_timeout_returns_504_and_stops_query$/) | test(/^reporting_usage::usage_summary::usage_summary_database_timeout_returns_504$/))'
+filter = 'package(api) & binary(e2e_all) & (test(/^database_encryption::/) | test(/^prepared_statement_cache::/) | test(/^reporting_usage::usage_export::usage_export_database_timeout_returns_504_and_stops_query$/) | test(/^reporting_usage::usage_summary::usage_summary_database_timeout_returns_504$/))'
 test-group = "e2e-db"
 threads-required = "num-test-threads"
 

crates/api/tests/e2e_all/main.rs

@@ -67,6 +67,7 @@ mod org_system_prompt;
 mod organization_deletion;
 mod pagination_validation;
 mod patroni_failover;
+mod prepared_statement_cache;
 mod privacy_classify;
 mod privacy_redact;
 mod provider_errors;