shard_reader: Continue after read_ahead error

If read ahead failed, just issue a log warning
and proceed to close the reader.

Currently co_await will throw and the evictable reader
won't be closed.

This is seen occasionally in testing, e.g.
https://jenkins.scylladb.com/view/master/job/scylla-master/job/dtest-debug/1010/artifact/logs-all.debug.2/1640918573898_lwt_banking_load_test.py%3A%3ATestLWTBankingLoad%3A%3Atest_bank_with_nemesis/node2.log
```
ERROR 2021-12-31 02:40:56,160 [shard 0] mutation_reader - shard_reader::close(): failed to stop reader on shard 1: seastar::named_semaphore_timed_out (Semaphore timed out: _system_read_concurrency_sem)
```

Fixes #9865.

Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
Message-Id: <20220102124636.2791544-1-bhalevy@scylladb.com>

mutation_reader.cc

@@ -1645,11 +1645,15 @@ future<> shard_reader::close() noexcept {
         co_return;
     }
 
-    try {
-        if (_read_ahead) {
+    if (_read_ahead) {
+        try {
             co_await *std::exchange(_read_ahead, std::nullopt);
+        } catch (...) {
+            mrlog.warn("shard_reader::close(): read_ahead on shard {} failed: {}", _shard, std::current_exception());
         }
+    }
 
+    try {
         co_await smp::submit_to(_shard, [this] {
             auto irh = std::move(*_reader).inactive_read_handle();
             return with_closeable(flat_mutation_reader(_reader.release()), [this] (flat_mutation_reader& reader) mutable {