feat: add get_experiment_groups() to expose group names and return values

[cam-matsui]

Why?

When building generic experimentation into a "platform", one must be cautious in deciding if and when to log exposures to avoid dilution. For example:

We have a rules engine and we want to generically provide the ability to experiment between 2 rules that evaluate whether a user gets an email. We might set up an experiment like:

Control	Test
rule_id=A	rule_id=B

One option is just to call get_experiment(user) and evaluate the rule with the returned ID. However, if rule A and rule B both return the same result (e.g., false) for some subset of users, we will dilute our experiment by exposing those users into the experiment even though none of those users would receive the email either way.

In most cases, we should instead evaluate both rules for all users in the experiment, and only log exposure if the results (and therefore the user experience) differ for rule A and B. In these kinds of scenarios, we need access to configurations for all experiment groups.

The only ways to achieve this today are to:
a) use the console client to make an HTTP request, which is redundant as we already have this data in-memory
b) reach into the internals of the SDK to pull this data, which adds dependencies on interfaces that may change.

Because we have all of this data available in memory anyways, we are proposing to extend the interface to avoid the above workarounds.

Summary

Adds a new public get_experiment_groups function that returns the group name and return value for each group in a given experiment, without requiring a user evaluation.

New API

# On StatsigServer instance
groups = server.get_experiment_groups("my_experiment")

# Module-level
groups = statsig.get_experiment_groups("my_experiment")

Each entry in the returned list has the shape:

{
    "group_name": "Control",
    "return_value": {"button_color": "blue", ...}
}

Behavior

• Returns [] if the experiment does not exist or the name refers to a dynamic config (not an experiment)
• Excludes rules explicitly marked with isExperimentGroup: false (holdout/sizing rules on production specs)
• Wrapped in the standard _errorBoundary.capture pattern, consistent with all other public methods

Changes

• statsig/statsig_server.py — StatsigServer.get_experiment_groups
• statsig/statsig.py — module-level get_experiment_groups
• tests/test_get_experiment_groups.py — 11 tests covering both the server method and the module-level function