[WIP] Refactor allFit to avoid update() in optimizer refits

[Copilot]

This PR refactors allFit() to avoid update() in the optimizer refit loop and rely on the more direct refit() path instead.

Changes made

• Updated R/allFit.R so allFit() now refits each optimizer run via refit(object, control=ctrl) rather than update(...).
• Removed the remaining internal update() fallback from the allFit refit path.
• Preserved optimizer/method control handling (optCtrl, maxfun, optimizer-specific method/algorithm wiring).
• Kept the nlmer warning behavior guarded by start_from_mle && isNLMM(object).
• Updated the focused regression test in tests/testthat/test-allFit.R to validate robustness without relying on update()-style call$data lookup.

Validation

• Attempted targeted test execution for test-allFit, but this environment does not provide R/Rscript, so runtime tests could not be executed locally.
• Performed static/diff validation and automated code review/security checks in this environment.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>refactor allFit</issue_title>
<issue_description>allFit uses update(), which is fragile and slower than it needs to be. It would be faster and more robust to only re-do the optimize(Gl|L)mer + mkMermod part of the pipeline ... More generally, update() could be optimized to be smarter about which components are being changed and only re-running the necessary bits: i.e., generalizing from refit() (only response variable changes) to use cases like

• fixed effects change, structure of random effects doesn't (can substitute new X and leave everything else alone)
  </issue_description>

<agent_instructions>Basically, try to find a way to make the allFit function faster.</agent_instructions>

Comments on the Issue (you are @copilot in this section)

• Fixes refactor allFit #12

---

📍 Connect Copilot coding agent with Jira, Azure Boards or Linear to delegate work to Copilot in one click without leaving your project management tool.

[annahuynhly]

@copilot could you please try again?

[Copilot]

@copilot could you please try again?

Implemented a new attempt and pushed it in 70d31d6: allFit(start_from_mle=TRUE) now takes a faster/less-fragile path via refit(object, control=ctrl) (with NLMM fallback preserving prior update(..., start=...) behavior), while start_from_mle=FALSE keeps existing update behavior. I also added a focused regression test for start_from_mle data-lookup robustness.

[annahuynhly]

@copilot please really focus on this key comment:

It would be faster and more robust to only re-do the optimize(Gl|L)mer + mkMermod part of the pipeline ... More generally, update() could be optimized to be smarter about which components are being changed and only re-running the necessary bits:

[annahuynhly]

@copilot don't you think that it's ironic you used update here, but the main point is to NOT use update...?

[Copilot]

Addressed in b0e5dae (plus e950e49): I removed the remaining update() usage from allFit() and switched the refit path to refit(object, control=ctrl) for optimizer runs. No UI changes here, so there is no UI screenshot.

[Copilot]

@copilot please really focus on this key comment:

It would be faster and more robust to only re-do the optimize(Gl|L)mer + mkMermod part of the pipeline ... More generally, update() could be op...

Done in b0e5dae (with follow-up e950e49): allFit() now avoids update() in its refit loop and uses refit(object, control=ctrl) so it re-runs optimizer/model rebuild without the full update pipeline. I also updated the robustness test accordingly. No UI changes in this PR, so there is no UI screenshot.