Impossible to cancel replications in some scenarios

In some scenarios it's impossible to cancel a replication by posting to /_replicate, namely:

1) A filtered replication is started, the filter's code is updated in the source database, therefore's a subsequent cancel request will not generate the old replication ID anymore, has it got a different filter code;

2) Dynamically changing the httpd port will also result in the impossibility of computing the right replication ID

Finally, it's also nicer for users to not need to remember the exact replication object posted before to /_replicate.
The new approach, in addition to the current approach, allows something as simple as:

POST /_replicate

{"replication_id": "0a81b645497e6270611ec3419767a584+continuous+create_target", "cancel": true}

The replication ID can be obtained from a continuous replication request's response (field "_local_id"), _active_tasks (field "replication_id") or from the log.
Aliases "_local_id" and "id" are allowed instead of "replication_id".

Patch:

https://github.com/fdmanana/couchdb/commit/c8909ea9dbcbc1f52c5c0f87e1a95102a3edfa9f.diff

(depends on COUCHDB-1266)