BOT: Fix #983: Create .data.frame methods for as_forecast_* generics

[nikosbosse]

CLAUDE:

Summary

• Renames the 8 existing .default methods for as_forecast_*() generics to .data.frame methods, since they assume data.frame-like input
• Creates new .default methods that error with a clear message when non-data.frame input is provided (e.g. vectors, matrices, lists)
• Adds tibble to Suggests for test coverage of tibble dispatch

Fixes #983

Root cause

All as_forecast_*() generics only had .default methods which silently accepted any type via as.data.table() coercion in ensure_data.table(). This meant vectors, matrices, and other non-data.frame types would either silently produce nonsensical results or fail with confusing deep validation errors.

What the fix does

• Renamed .default → .data.frame for all 8 forecast types (binary, point, sample, quantile, nominal, ordinal, multivariate_sample, multivariate_point)
• Created new .default methods that immediately error with: "Input must be a data.frame or similar (e.g. a data.table or tibble), not <class>."
• S3 dispatch ensures data.table and tibble inputs still route to the .data.frame method since both inherit from data.frame

Test coverage added

• Tests covering:
  • Plain data.frame input works for all types (regression guards)
  • data.table input still dispatches correctly via S3
  • Non-data.frame input (vectors, matrices, lists, strings) produces clear errors for all types
  • Column renaming functionality preserved after method rename
  • tibble input dispatches correctly

Test plan

• All 789 tests pass (0 failures, 0 warnings)
• Linter passes
• Codecov patch and project checks pass

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.00%. Comparing base (4a222da) to head (515abd3).
⚠️ Report is 26 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1077      +/-   ##
==========================================
+ Coverage   97.98%   98.00%   +0.02%     
==========================================
  Files          37       37              
  Lines        1984     2008      +24     
==========================================
+ Hits         1944     1968      +24     
  Misses         40       40              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[nikosbosse]

CLAUDE: Clean, well-structured PR. Correctly renames .default to .data.frame across all 7 as_forecast_*() generics and adds new .default methods that error with clear messages for non-data.frame input. S3 dispatch is correctly leveraged for data.table and tibble inheritance. NAMESPACE, documentation, and tests are all consistent. 12 new tests cover regression guards, dispatch correctness, error handling, column renaming, and tibble compatibility. Two very minor cosmetic notes: (1) as_forecast_sample.data.frame is missing the explicit @method roxygen tag (inferred correctly, no functional impact), (2) the data.table dispatch test covers 6/7 types (missing multivariate_sample, very low risk). Verdict: approve.

[nikosbosse]

@seabbs it's not a great PR, but it's not a bad PR either. I have no strong attachments to it, but I think it's fine to merge

[seabbs]

Isn't this a breaking change?

[seabbs]

I think as written this is a breaking change as the default input goes from anything that be coerced into a data.table into anything that is a data.frame. That should at the minimum require changes to the documentation (which should say above). Do we have user reports saying this is an issue etc etc. I think not and based on that I wonder if we should just leave as is, perhaps check to make sure the docs are clear, and close the target issue?

[nikosbosse]

Huh. It didn't occur to me that there might be things you could coerce into a data.table that aren't a data.frame. But yeah, lists are famously that...
Good catch, thank you. Just didn't think of it. The PR is more meant to clean things up, I don't think there is any actual user need. Let shelve it.