Fix HdfsTimeoutException not retried in setupBlockReader and EC createBlockReader

[exmy]

HdfsTimeoutException is a sibling class of HdfsIOException (both extend
HdfsException), not a subclass. In three places where RemoteBlockReader
construction can throw HdfsTimeoutException (via checkResponse() →
readVarint32()), the exception was not caught by the existing
catch(HdfsIOException) blocks, causing the read to fail immediately
without retrying another DataNode.

This affects both replicated files (InputStreamImpl::setupBlockReader)
and Erasure Coded files (StripedInputStreamImpl::createBlockReader).
For EC files, the uncaught timeout causes chunks to be incorrectly
counted as "missing", leading to "Missing blocks, missingChunksNum >
parityBlkNum" errors even when the data is intact but DataNodes are
temporarily unresponsive.

Changes:

• InputStreamImpl::setupBlockReader (both overloads): catch
  HdfsTimeoutException, add timed-out node to failedNodes, and retry
  another DataNode.
• StripedInputStreamImpl::createBlockReader: same fix for the EC path.
• RemoteBlockReader::readPacketHeader: catch HdfsTimeoutException and
  wrap as HdfsIOException with Block/Datanode context, ensuring
  timeouts during data reading are also properly handled by upstream
  retry logic.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

xumingyong seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[zhanglistar]

LGTM @alexey-milovidov pls check.