Sending zero length search request causes infinite loop in casDGClient::processDG()

Reported by Shuei YAMADA shuei.yamada_AT_gmail.com:

Last april I posted to tech-talk an problem that cagateway runs away
on rare occasions (https://epics.anl.gov/tech-talk/2017/msg00714.php).
I successfully reproduced the problem by UDP-port scan with nmap.
When I run nmap as following:

nmap -sU -p 5064 -A ip.address.of.cagateway

- all PVs subscribing via cagateway become disconnected,
- cagateway is eating up the CPU,
- no distinguishable log messege, no "zero length PV name in UDP
search request?" either,
- excas shows the same simptom.

There is no way to exiting the the while() loop in
casDGClient::processDG() when program reaches at the end of while()
block with a condition such that:
- this->in.bytesPresent()>0 && dgInBytesConsumed == 0 && status ==
S_cas_success.

We are using base R3.14.12.3 for production and R3.15.5 for evaluation
at our site and both have the problem. Also R3.14.12.7 and R3.16.1
seem to have the same problem. Please find a naiive fix for this in
the attachment.