[vignesh2027]

Calling Write, Flush or SetNbWorkers on a Writer after Close crashes the whole process instead of returning an error.

Since #163, Close frees the C compression stream and sets w.ctx to nil, which fixed the leak and made double Close safe. The other Writer methods do not check for that, so they pass a nil ZSTD_CCtx to libzstd and the program dies with a SIGSEGV inside cgo. Because it happens in C code, recover cannot catch it.

var buf bytes.Buffer
w := zstd.NewWriter(&buf)
w.Write([]byte("hello"))
w.Close()
w.Write([]byte("again")) // SIGSEGV, signal arrived during cgo execution

The same thing happens after a Close that fails because the underlying writer returned an error, since that path also nils ctx without setting firstError.

This change makes Write, Flush and SetNbWorkers return a new errWriterClosed when ctx is nil, the same way reader.Read already returns errReaderClosed after Close. An earlier firstError is still returned first, and a second Close still returns nil, so the existing tests from #163 keep passing.

I added two tests next to the #163 ones. On main both crash the test binary with SIGSEGV in ZSTD_compressStream2_wrapper, and with the fix they pass. I ran go test with the CI payload, go test -race, and the suite with -tags external_libzstd against a system libzstd, all green.