Move scheduler HTTP API parsing out of master process

Today all calls sent to Mesos via the v1 api are parsed in the master process, even though non of the information contained in that process is required for this. The master process is already doing the majority of the work and is the biggest bottleneck, so anything that can be done elsewhere would improve the situation. Parsing of the calls (and some validation) could be done in separate processes and in parallel for multiple frameworks, which should improve overall responsiveness and throughput of the master process.