Fix #606: numerical correctness assertions for Combine.stack/take tests + optimize Combine.stack()

[Copilot]

Follow-up from #579 review. Two test gaps and one efficiency issue in Combine.

Test coverage gaps addressed:

• test_combine_stack — now splits the grouped array into two real partitions, fits transforms on each independently, stacks them, and asserts np.testing.assert_allclose(stacked.update(full_ga), concat([u1, u2])) to verify numerical correctness rather than just structure.

• test_nested_combine_take — now builds a subset_ga for the selected indices, fits a fresh transform on it, and asserts the take()-derived transform produces bit-identical update() output to the freshly-fitted one.

Efficiency fix:

Combine.stack() previously deepcopy(transforms[0]) and then immediately overwrote out.tfm1 / out.tfm2 — wasting the deep copy of those attributes. Fixed by creating a shallow stub with tfm1/tfm2 nullified before deepcopy, leaving the input object unmutated:

stub = copy.copy(first)
stub.tfm1 = stub.tfm2 = None
out = copy.deepcopy(stub)
out.tfm1 = first.tfm1.stack([tfm.tfm1 for tfm in transforms])
out.tfm2 = first.tfm2.stack([tfm.tfm2 for tfm in transforms])

Checklist:

• This PR has a meaningful title and a clear description.
• The tests pass.
• All linting tasks pass.
• The notebooks are clean.

Original prompt

Create a pull request that addresses all items described in GitHub issue #606 in the Nixtla/mlforecast repository:

• Review the issue description and any linked comments, code snippets, stack traces, or reproduction steps.
• Identify the root cause(s) in the codebase.
• Implement a robust fix and any necessary refactors.
• Add/extend automated tests to cover the reported behavior and prevent regressions.
• Update documentation/examples if the fix changes user-facing behavior.

Issue reference (must be used as the source of requirements): #606

Acceptance criteria:

• All requested changes from the issue are implemented.
• New/updated tests pass in CI.
• Code style and linting remain clean.
• Include a clear PR description that links to and closes issue Follow-ups from Combine distributed fix (yherin/fix-distributed-combine-transform) #606 (e.g., "Closes Follow-ups from Combine distributed fix (yherin/fix-distributed-combine-transform) #606").

The following is the prior conversation context from the user's chat exploration (may be truncated):

User: Open a pr that tackles the things described in here

Assistant: I’ve queued up the pull request creation request for Nixtla/mlforecast based on issue #606, and I’m waiting for you to accept/approve it in the dialog so I can continue.

This pull request was created from Copilot chat.

---

💬 Send tasks to Copilot coding agent from Slack and Teams to turn conversations into code. Copilot posts an update in your thread when it's finished.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
1 out of 2 committers have signed the CLA.

✅ nasaul
❌ Copilot
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[nasaul]

LGTM

[codspeed-hq]

Merging this PR will not alter performance

✅ 12 untouched benchmarks

---

_{Comparing copilot/fix-issue-606-implementation (9adbb69) with main (cb8b558)}