fix(MsSql): default-validate is now IsOpen, no per-acquire SELECT 1 round-trip

The previous DefaultValidateConnectionAsync ran "SELECT 1" against every
pooled connection on every acquire. Under cancellation pressure (HTTP
server forwarding requests whose CT can fire mid-flight), that round-trip
could stall on a connection whose previous response hadn't been fully
drained — taking seconds and inviting the CT to cancel mid-validation.
That was wedging CosmoS3 PUTs through cosmoproxy.

Drop validation to a cheap !IsOpen-or-true check and let WithRetryAsync's
existing dead-connection retry path detect actually-broken connections
when the real query throws. Operators who want a wire-level health probe
on every acquire can still pass their own validateConnection delegate.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: vkuttyp

src/CosmoSQLClient.MsSql/MsSqlConnectionPool.cs

@@ -331,24 +331,23 @@ private async Task<T> WithRetryAsync<T>(Func<MsSqlConnection, Task<T>> op, Cance
             }
         }
 
-        private async ValueTask<bool> DefaultValidateConnectionAsync(MsSqlConnection conn, CancellationToken ct)
-        {
-            if (!conn.IsOpen)
-                return false;
-            try
-            {
-                await conn.QueryAsync("SELECT 1", ct: ct).ConfigureAwait(false);
-                return true;
-            }
-            catch (OperationCanceledException)
-            {
-                throw;
-            }
-            catch
-            {
-                return false;
-            }
-        }
+        // Default validation is intentionally lightweight: just check IsOpen.
+        // We DO NOT run a "SELECT 1" round-trip on every acquire because
+        // — when the pool is under cancellation pressure (e.g. an HTTP server
+        // forwarding requests whose CT can fire at any moment) — that round-trip
+        // can stall on a connection whose previous response wasn't fully drained,
+        // leading to a request-time hang of seconds before the CT eventually
+        // cancels validation. WithRetryAsync already retries on dead-connection
+        // exceptions thrown by the real query (IOException, SocketException,
+        // SqlException with connection-error kind, etc.), so a stale-but-IsOpen
+        // connection will still recover transparently — it just defers the
+        // detection to the actual statement instead of spending a network
+        // round-trip on every acquire.
+        //
+        // Callers that need stronger validation can supply their own
+        // validateConnection delegate via the ctor.
+        private static ValueTask<bool> DefaultValidateConnectionAsync(MsSqlConnection conn, CancellationToken ct)
+            => new ValueTask<bool>(conn.IsOpen);
 
     // ── ISqlDatabase ───────────────────────────────────────────────────────────