Refine spot instance startup

[ansoncfit]

When a regional analysis triggers a request to start spot instance workers, if the request would lead the total number of workers to exceed our configured limit (~6x higher on prod vs. staging), the request is totally ignored.

This PR changes that behavior and reduces the likelihood requests will be ignored:

• Reduces the number of instances requested for higher zoom and non-transit analyses
• Adds a backoff function (h/t Zeno's paradox) -- if we are halfway to the limit on the total number of workers, no more than half of the available remaining slots will be requested

Three additional refinements could be considered -- per-access group limits, further reducing the number of instances requested when analyses don't have frequency-based routes (we don't yet expose hasFrequency() in a convenient place though), and sizing the spot instance request based on observed tasks per minute.

[abyrd]

Looks good to avoid the problem we've seen where way too many workers are created for a simple walk search, flooding the backend with responses from a single job. Longer term we'll want to more actively manage the worker pool using the components that were already drafted for that purpose.