Honor parameter distribution in Skopt and Goptuna suggestion services

[saivedant169]

What this PR does / why we need it

The Skopt and Goptuna suggestion services built their search spaces from parameter.type only and never read parameter.distribution, so a declared distribution was silently discarded:

• Skopt (skopt/base_service.py) hardcoded a "log-uniform" prior for every double parameter and left integer parameters uniform, so a distribution: uniform double (the default) was sampled log-uniformly.
• Goptuna (goptuna/converter.go) always built UniformDistribution / IntUniformDistribution, so a distribution: logUniform parameter was sampled uniformly.

Both now honor the declared distribution. Because both libraries support uniform and log-uniform directly (skopt.space.Real/Integer take a prior, and goptuna has LogUniformDistribution), the distribution is mapped through rather than rejected. For normal / logNormal, which neither library samples natively (and logUniform for goptuna int parameters, which goptuna has no distribution for), the service now raises a clear error naming the parameter and the unsupported distribution, consistent with the Optuna service in #2666.

Note this changes the Skopt default for double parameters from log-uniform to uniform, which is the intended correction.

Which issue(s) this PR fixes

Fixes #2688

Testing

• Added Goptuna unit tests in converter_test.go (log-uniform maps to LogUniformDistribution; normal on a double and logUniform on an int return an error). All pass with go test ./pkg/suggestion/v1beta1/goptuna/.
• Added Skopt unit tests in test_skopt_service.py covering the distribution to prior mapping and the fail-loud path for unsupported distributions. These run under make pytest-skopt.
• Verified the Skopt mapping against the actual create_optimizer: a uniform double now yields prior="uniform", a log-uniform double yields prior="log-uniform", and a normal double raises ValueError.

[saivedant169]

Hi @andreyvelich, whenever you get a window, mind taking a look at this one? Same distribution-handling line as #2666, just extended to the skopt and goptuna services, which were silently ignoring the declared distribution. No rush.

[andreyvelich]

Thanks @saivedant169!
/ok-to-test
/lgtm
/approve

[google-oss-prow]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: andreyvelich

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Details

Needs approval from an approver in each of these files:

• OWNERS [andreyvelich]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[saivedant169]

/retest

[coveralls]

Coverage is 9.986% — saivedant169:fix/skopt-goptuna-distribution into kubeflow:master. No base build found for kubeflow:master.

[saivedant169]

Hi @andreyvelich, this one has been approved and is green apart from the tune-example-llm-optimization e2e, which times out with the trial still Running rather than failing, so it looks unrelated to this change. A /retest last week hit the same timeout. Could you re-run it when you get a moment, or let me know if there is anything you would like me to adjust on my end?

[saivedant169]

@andreyvelich this one's still sitting on the flaky e2e, logs are from July. Mind kicking off a rerun?

[saivedant169]

/retest