Add support for environment and dataset rules with same names

[MikeGoldsmith]

Which problem is this PR solving?

Adds support for config files to contain both a dataset and environment with the same name in the same rules config. This is achieved by adding a DatasetPrefix to the config and then using that prefix when setting up rule definitions.

config.toml

DatasetPrefix = "dataset"

rules.toml

Sampler = "DeterministicSampler"
SampleRate = 1

[production] # environment
  Sampler = "DeterministicSampler"
  SampleRate = 10

[dataset.production] # dataset
  Sampler = "DeterministicSampler"
  SampleRate = 20

• Closes Add default environment prefix to allow environments and dataset with same name in rules file #408

Short description of the changes

• Adds new config option DatasetPrefix, defaults to ""
• If a legacy is provided to the SamplerFactory, it uses the configured prefix along with the sample key to locate rules definitions
• Adds test to verify behaviour

[kentquirk]

Generally approved, but two notes:

• I know there's probably going to be a docs-related PR for this as well, but I think it also deserves a note in the README.
• It took me a while to understand the description of this PR. Can you please add a bit of a note as to why there's a problem that needs this fix?

[seh]

You could mandate that the intervening period be part of the specified prefix. If I set the DatasetPrefix field to "x," the prefix really becomes "x."—even though I didn't mention the period when specifying my preferred prefix. I might prefer a prefix like "dataset-" (with a separating hyphen instead of a period).

Had you considered including this separator as part of the prefix? Were you attempting to avoid false aliasing if there was no obvious separator character included in the prefix?

[MikeGoldsmith]

Yes, I agree additional docs on the prefix format where the . separator is added for you would be useful 👍🏻

The separator is not necessary but I think helps create a mental divide. The intent of this change is to support users migrate from legacy datasets to newer environment and services rule sets and should not be a permanent solution. I think a simple solution of always using a single . that's well documented should be good enough.

[kentquirk]

Thanks for the docs improvement. I added suggestions for a couple of typos.