fix: close tmp key file in fuzz test (ethereum#29444)

[hmlee-wm]

Summary

Ports ethereum#29444. createKeyPair() opened a temp file via os.CreateTemp but never closed it. Other temp files in the same file (tmpFile, pubKeyFile) already have defer Close(). Add the missing defer tmpKey.Close().

Upstream References

PR	Description
#29444	crypto/signify: close tmp key file in test

Changes

• crypto/signify/signify_fuzz.go: add defer tmpKey.Close() in createKeyPair()

[0xmhha]

Note: the fuzz harness in this file cannot be built, so the change is unverifiable

No impact on the normal build or make test — signify_fuzz.go carries
//go:build gofuzz, so it is excluded from both. go build ./crypto/signify/
passes. The fuzzer is also not registered in oss-fuzz.sh, so nothing in CI
builds it either.

The problem only surfaces when the harness is built with the tag:

$ go build -tags gofuzz ./crypto/signify/
# github.com/ethereum/go-ethereum/crypto/signify
crypto/signify/signify_fuzz.go:59:8: undefined: SignifySignFile
crypto/signify/signify_fuzz.go:71:9: no new variables on left side of :=

go vet reports the same, but stops at the first error — worth knowing, since it
makes the problem look like a single-line fix when it is not:

$ go vet -tags gofuzz ./crypto/signify/
# github.com/ethereum/go-ethereum/crypto/signify
# [github.com/ethereum/go-ethereum/crypto/signify]
vet: crypto/signify/signify_fuzz.go:59:8: undefined: SignifySignFile

Both errors are pre-existing and reproduce on dev:

1. L59 — SignifySignFile was renamed to SignFile in f935b1d
   (ethereum#21977); the call site was never updated. It is the only remaining
   reference in the tree.
2. L71 — _, err := exec.LookPath(signify); err is already declared at L37
   and _ is not a new variable, so := has nothing new to declare.

Upstream fixed both in ethereum#33402 (a9eaf2ffd859, 2025-12-13). Applying that
patch locally makes go build -tags gofuzz ./crypto/signify/ pass.

The change in this PR is correct on its own: defer tmpKey.Close() is registered
after the three os.Remove defers, so LIFO order closes the handle before the
files are removed.

[colinkim]

LGTM.
Since the gofuzz-tagged harness build failure predates this PR, and CI does not exercise that tag, how about tracking it in a separate follow-up issue or PR?