fix: the cpuload 100% in case the tcpstream WouldBlock

[garyyu]

Fix #3025

In #2855, we refactored the p2p code to use blocking IO instead of non-blocking, which returns us the dramatically reduced CPU load.

But the Rust TcpStream behavior for blocking mode is not as expected in some cases.

For example, the stream.read_exact() will practically return immediately if the underlined stream would block.

i.e. the so-called WouldBlock error:
Err(Os { code: 35, kind: WouldBlock, message: "Resource temporarily unavailable" })

Then, the p2p poll() will just seamlessly call the stream.read_exact() in the loop, which in result will give us a 100% cpu load at this case.