BOT: Fix #895: Add get_non_monotonic_forecasts() and monotonicity warning

[nikosbosse]

Summary

• Adds a new exported get_non_monotonic_forecasts() diagnostic function (modeled on get_duplicate_forecasts()) that identifies forecast units where quantile predictions are not monotonically non-decreasing with increasing quantile levels
• Adds a warning in assert_forecast.forecast_quantile() (guarded by verbose) so users are alerted at as_forecast_quantile() time rather than only discovering the issue deep inside bias_quantile() during scoring
• Also addresses Add check to validate_forecast() that predictions must not decrease with quantile #412 (add monotonicity check to forecast validation)

Root cause: Non-monotonic quantile predictions silently pass through as_forecast_quantile() and assert_forecast(). The only check exists deep in bias_quantile_single_vector() as a hard error, which users only encounter at score() time — often after significant pipeline setup.

What changed:

• New R/get-non-monotonic-forecasts.R with exported get_non_monotonic_forecasts() and internal check_monotonicity()
• assert_forecast.forecast_quantile() now calls check_monotonicity() and emits a cli_warn() when violations are found
• Warning is suppressible via verbose = FALSE
• 15 new tests in test-get-non-monotonic-forecasts.R + 4 new tests in test-class-forecast-quantile.R

Closes #895
Closes #412

Test plan

• New get_non_monotonic_forecasts() tests: well-formed data returns empty, detects violations, mixed units, counts argument, custom forecast_unit, class checks, data.frame input, equal predictions edge case
• check_monotonicity() tests: returns TRUE for good data, message string for bad data
• assert_forecast.forecast_quantile() tests: warns on non-monotonic, no false positive, verbose=FALSE suppresses, end-to-end score works with bias excluded
• Full test suite passes (699 tests, 0 failures)
• R CMD check: 0 errors, 0 warnings, 2 notes (pre-existing)

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.86%. Comparing base (ac0c01a) to head (7f00d40).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1081      +/-   ##
==========================================
+ Coverage   97.83%   97.86%   +0.03%     
==========================================
  Files          35       36       +1     
  Lines        1845     1875      +30     
==========================================
+ Hits         1805     1835      +30     
  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: Review verdict — APPROVE. Clean implementation following the established get_duplicate_forecasts()/check_duplicates() pattern. Core detection logic (diff(predicted) < 0) correctly flags strictly decreasing predictions while allowing equal values. Warning (not error) in assert_forecast.forecast_quantile() is the right design choice per issue discussion. All 12 test specifications covered by 15 tests. No issues found.