fix: delete leftover heartbeat connections

[df-wg]

Users reported seeing the below sigfault when a connection to a subgraph is interrupted over multipart:

cosmo-router     | 21:07:16 PM ERROR core/graphql_handler.go:380 Unable to write error response {"hostname": "ad1991cdcfd2", "pid": 1, "component": "@wundergraph/router", "service_version": "0.158.0", "request_id": "ad1991cdcfd2/RTnCLg1J8M-000022", "trace_id": "77476c6fe295cd7dafeb71e00183879b", "error": "context canceled"}
cosmo-router     | github.com/wundergraph/cosmo/router/core.(*GraphQLHandler).WriteError
cosmo-router     |      github.com/wundergraph/cosmo/router/core/graphql_handler.go:380
cosmo-router     | github.com/wundergraph/graphql-go-tools/v2/pkg/engine/resolve.(*Resolver).handleHeartbeat.func1
cosmo-router     |      github.com/wundergraph/graphql-go-tools/v2@v2.0.0-rc.136/pkg/engine/resolve/resolve.go:431

After investigating, the cause seemed to be a number of times where we deleted the subscription trigger but didn't clean up the heartbeat (which is running in a separate thread), causing it to write on a non-existent context. This PR cleans that up

[StarpTech]

Great catch, looks reasonable to me. Can we add a test for it here or in the router?

[Noroth]

Looks good. Do you think we should have one more test with 50 subscriptions at once and make sure that all of them are deleted properly?

e.g.:
require.Eventually

[df-wg]

Good point @Noroth , added a test like that