Increase MSCCLPP_BITS_REGMEM_HANDLE to 9 (#251)

MSCCLPP_BITS_REGMEM_HANDLE=8 limits the number registered memories for a
ProxyService to 256. Many use cases, such as KV cache transfer, require
registering more tensors.

This change allows registering up-to 512 memories. Note that this change
uses up the slack bits remaining in the ChannelTrigger struct.

include/mscclpp/proxy_channel_device.hpp

@@ -22,7 +22,7 @@ const TriggerType TriggerSync = 0x4;  // Trigger a flush.
 
 #define MSCCLPP_BITS_SIZE 32
 #define MSCCLPP_BITS_OFFSET 32
-#define MSCCLPP_BITS_REGMEM_HANDLE 8
+#define MSCCLPP_BITS_REGMEM_HANDLE 9
 #define MSCCLPP_BITS_TYPE 3
 #define MSCCLPP_BITS_CONNID 10
 #define MSCCLPP_BITS_FIFO_RESERVED 1