add monitoring.ignore-duplicates to skip duplicate time series

[hcavarsan]

Problem

the Cloud Monitoring API sometimes returns the same time series more than once in a single scrape, and prometheus refuses the duplicate while gathering:

collected metric "stackdriver_..." { ... } was collected before with the same name and label values 

promhttp turns that into a 500/503, so one duplicate fails the whole scrape.

Solution

this adds monitoring.ignore-duplicates (off by default). when it's on, the gatherer drops the duplicate-series error and returns what was collected.

only exact duplicates get dropped, since prometheus decides what counts as the same series by name, labels, and timestamp. anything differing in a label or timestamp is kept, and with the flag off nothing changes.

Testing

collectors/gatherer_test.go has an end-to-end test where a duplicate gets suppressed and the series collapses. it also has a table test for the error filtering.

related: #57, #103, #166, #450, #470.