Sample more for the initial days of release

[ghost]

Upon new releases of Firefox, the HB score is too noisy due to throttling / updates being slow on initial rollout resulting in a smaller sample of users.

Proposed solution for heartbeat-by-user-first-impression for the Release channel (for now):

1. Add a sampleMultiplier function that considers the version, channel, and current date to provide an additional multiplier to the sample rate. This will be utilized in shouldRun.
2. sampleMultiplier will utilize a manually curated (for now) table to look up the release date for the version of the client. We will use date math to see how many days it has been since the release date. Let's call this diff daysSinceRelease for now
3. We will use daysSinceRelease to calculate the sample multiplier based on historical rollout data. This may be a heuristic formula or a lookup table

Notes

• This feature cannot leak ADI, we should only be dealing with multipliers
• This feature is only release
• The multiplier should probably be bounded
• All our lookups need to be sparse data safe (i.e. we will not curate multipliers for all permutations of our data)

Rob will make a first attempt on this. Gregg will review the code. Ilana and Gregg will review the heuristic.

[MattGrimes]

Solid. Do you have an ETA on when you might have the first attempt done?

[ghost]

I may be being overly optimistic, but I don't think that this is a difficult project. I'd say we could test for 42 without issue.

[gregglind]

Spec clarifications:

• We are only concerned with sample by 'series', so 42 is 42.*. If we have to get fancy about point releases we will revisit.