Potential memory leak/bug in the valkey client

[jskrzypek]

I think there might be an issue with the __aenter__ method on the ValkeyBackend.

aiocache/aiocache/backends/valkey.py

Lines 32 to 37 in d685fff

	async def __aenter__(self) -> Self:
	self.client = await GlideClient.create(self.config)
	return self
	async def __aexit__(self, *args, **kwargs) -> None:
	await self.client.close()

This code snippet explains my thinking.

async def cache_operations():
    client_1 = None
    async with ValkeyCache(
        config=GlideClientConfiguration(addresses=[NodeAddress("localhost", 6379)])
    ) as my_cache_1: # <<< new client created
        client_1 = my_cache_1.client
        async with my_cache_1 as my_cache_2: # <<< new client created
            assert my_cache_1 is my_cache_2  # because of return self
            client_2 = my_cache_2
            assert client_1 != client_2
            assert my_cache_1.client == client_2
            # exit inner with, call __aexit__, calls await self.client.close(); i.e. client_2.close())
        assert my_cache_1.client != client_1 # it's still client_2
        # exit inner with, call __aexit__, calls await self.client.close(); i.e. client_2.close() again, which may error.
    # client_1.close() may never be called

I think that the __aenter__ method should maybe handle checking if an existing connection is available and disposing of it if not, or else return a clone of self with a new client to ensure that the connection does get managed correctly.

[Dreamsorcerer]

A user shouldn't be nesting the same context multiple times. Feel free to create a PR that adds a check that throws a RuntimeError.