add required locking (mutex) to GetMempoolTx

[LarryRuane]

Closes #402

To reproduce the problem, call the GetMempoolTx gRPC concurrently. For example, run the following in two shell windows simultaneously:

while :;do grpcurl -plaintext localhost:9067 cash.z.wallet.sdk.rpc.CompactTxStreamer/GetMempoolTx;done >/dev/null

The lightwalletd instance should panic within several seconds.

Exception has occurred: panic
"runtime error: invalid memory address or nil pointer dereference"
Stack:
	 4  0x0000000000dc387a in github.com/zcash/lightwalletd/frontend.(*lwdStreamer).GetMempoolTx
	     at /g/lightwalletd/frontend/service.go:479
	 5  0x0000000000d9248a in github.com/zcash/lightwalletd/walletrpc._CompactTxStreamer_GetMempoolTx_Handler
	     at /g/lightwalletd/walletrpc/service_grpc.pb.go:632

The problem is that mempoolMap may have been changed (by another thread) during the current thread's single execution of this function, so that when it's indexed by txid at line service.go:479, that key may not exist.

I looked for other (unreported) instances of this problem and didn't find any, although did find that GetLatestBlock() has a "logical" race condition; it may return a height and hash that don't correspond to each other since they're fetched separately without locking. So this PR also fixes this problem.

[pacu]

utAck

[LarryRuane]

Force-pushed rebased