Not precisely an issue of the evolution, as it uses to be with workarounds,
but better to catch it here, when it's possible. This avoids crash of
the evolution on quit of the second instance and some critical warnings,
while still not leaking the EShell instance.