rpc: update win-library + use atomics

[holiman]

The windows named-pipe implementation hasn't been kept up to date for the last seven years, this PR switches it over to a somewhat more active fork. I want to see if it builds successfully on windows/386, which most often is failing with errors like

panic: The handle is invalid.
goroutine 10385 [running]:
internal/poll.execIO(0xa5eedd4, 0xce9620)
	C:/Users/appveyor/AppData/Local/geth-go-1.20.6-windows-amd64/go/src/internal/poll/fd_windows.go:201 +0x44f
internal/poll.(*FD).Read(0xa5eedc0, {0xb1db000, 0x1000, 0x1000})
	C:/Users/appveyor/AppData/Local/geth-go-1.20.6-windows-amd64/go/src/internal/poll/fd_windows.go:436 +0x13b
net.(*netFD).Read(0xa5eedc0, {0xb1db000, 0x1000, 0x1000})
	C:/Users/appveyor/AppData/Local/geth-go-1.20.6-windows-amd64/go/src/net/fd_posix.go:55 +0x3f
net.(*conn).Read(0xb199728, {0xb1db000, 0x1000, 0x1000})
	C:/Users/appveyor/AppData/Local/geth-go-1.20.6-windows-amd64/go/src/net/net.go:183 +0x4f
net/http.(*persistConn).Read(0xa9e2780, {0xb1db000, 0x1000, 0x1000})
	C:/Users/appveyor/AppData/Local/geth-go-1.20.6-windows-amd64/go/src/net/http/transport.go:1943 +0x141
bufio.(*Reader).fill(0xa687170)
	C:/Users/appveyor/AppData/Local/geth-go-1.20.6-windows-amd64/go/src/bufio/bufio.go:106 +0xe9
bufio.(*Reader).Peek(0xa687170, 0x1)
	C:/Users/appveyor/AppData/Local/geth-go-1.20.6-windows-amd64/go/src/bufio/bufio.go:144 +0x6d
net/http.(*persistConn).readLoop(0xa9e2780)
	C:/Users/appveyor/AppData/Local/geth-go-1.20.6-windows-amd64/go/src/net/http/transport.go:2107 +0x1df
created by net/http.(*Transport).dialConn
	C:/Users/appveyor/AppData/Local/geth-go-1.20.6-windows-amd64/go/src/net/http/transport.go:1765 +0x1480
FAIL	github.com/ethereum/go-ethereum/rpc	14.792s

and

ok  	github.com/ethereum/go-ethereum/rlp/rlpgen	3.318s
runtime.preemptM: duplicatehandle failed; errno=6
fatal error: runtime.preemptM: duplicatehandle failed
runtime stack:
runtime.throw({0x132f707?, 0xffffffffffffffff?})
	C:/Users/appveyor/AppData/Local/geth-go-1.20.6-windows-amd64/go/src/runtime/panic.go:1047 +0x65 fp=0xd5731ff870 sp=0xd5731ff840 pc=0xe6a7e5
runtime.preemptM(0xd5731ffdf0?)
	C:/Users/appveyor/AppData/Local/geth-go-1.20.6-windows-amd64/go/src/runtime/os_windows.go:1345 +0x4a5 fp=0xd5731ffdd0 sp=0xd5731ff870 pc=0xe67f85
runtime.preemptone(0xc006023800?)
	C:/Users/appveyor/AppData/Local/geth-go-1.20.6-windows-amd64/go/src/runtime/proc.go:5572 +0x5e fp=0xd5731ffde8 sp=0xd5731ffdd0 pc=0xe788fe
runtime.preemptall()
	C:/Users/appveyor/AppData/Local/geth-go-1.20.6-windows-amd64/go/src/runtime/proc.go:5534 +0x55 fp=0xd5731ffe20 sp=0xd5731ffde8 pc=0xe78855
runtime.forEachP(0x1358c50)
	C:/Users/appveyor/AppData/Local/geth-go-1.20.6-windows-amd64/go/src/runtime/proc.go:1676 +0xdb fp=0xd5731ffe88 sp=0xd5731ffe20 pc=0xe6fe7b
runtime.gcMarkDone.func1()
	C:/U

[holiman]

CI failure for Image: Visual Studio 2019; Environment: GETH_ARCH=386, GETH_MINGW=C:\msys64\mingw32 was

ok  	github.com/ethereum/go-ethereum/eth/catalyst	5.986s
--- FAIL: TestSkeletonSyncRetrievals (2.73s)
    skeleton_test.go:908: test 6, mid state: dropped peers mismatch: have 0, want 1
FAIL
FAIL	github.com/ethereum/go-ethereum/eth/downloader	162.714s

But the rpc tests passed. Rerunning it now

[fjl]

I don't think these failures are related to the named pipe library.

[holiman]

I'll re-run the appveyor builds, and update the list below with the status for the 386 build

1. RPC Success (failed on unrelated)
2. Success (no fails )
3. Success
4. Success
5. Success (failed on TestSkeletonSyncRetrievals)
6. Success

[holiman]

I don't think these failures are related to the named pipe library.

The reason I suspect they are, is that e.g this panic does arise in the windows file descriptor polling.

C:/Users/appveyor/AppData/Local/geth-go-1.20.6-windows-amd64/go/src/internal/poll/fd_windows.go:201

[fjl]

After looking into the new npipe depency code, I think it probably could have an effect on this error after all. But seems weird to add this dependency.

Alternative to explore: https://github.com/microsoft/go-winio

[holiman]

closing in favour of #27972