Fix monthly tasmax/tasmin being a copy of tas (#644)

[rbeucher]

Closes #644

The problem

Monthly tasmax and tasmin (CMIP7 tmaxavg/tminavg) were written with exactly the same values as tas. The daily variants were also wrong: they contained the daily mean temperature, not the daily max/min.

Why it happened

The mapping pointed all these variables at fld_s03i236, which is the mean temperature field. The real daily max/min live in separate fields (fld_s03i236_max / fld_s03i236_min) that only exist in the daily output stream — the monthly stream doesn't have them at all.

So for monthly targets, MOPPy read the monthly-mean field and asked for the "monthly maximum" of it. With one value per month, that operation returns the input unchanged — a silent copy of tas. For daily targets, it simply renamed the daily-mean field.

The fix

1. Read the right fields: tasmax/tasmin now read fld_s03i236_max/_min from the daily files (ESM1-5 and ESM1-6 mappings), even for monthly output.
2. Do the right reduction: monthly values are now the mean over days of the daily max/min (new calculate_monthly_mean operation), matching the CF definition "time: maximum within days time: mean over days". Daily output just renames the true extrema field.
3. Fail loudly next time: the monthly reduction functions now raise an error if fed monthly data instead of silently copying it, and the resampling-method detection correctly handles chained cell_methods (the substring-ordering bug described in [BUG] CMIP7 tmaxavg/tminavg output is a copy of tavg, not a daily max/min aggregate #644, which also affects e.g. sfcWindmax).
4. The auto-resampler now leaves the input alone when the variable's own formula does the time reduction, so the formula receives the raw daily data.

Verification

Against the ESM1.6 test archive (1951):

output	before	after	expected
tmaxavg (mon)	277.70 (= tas)	279.57	279.57
tminavg (mon)	277.70 (= tas)	275.82	275.82
tmax (day)	277.71 (= daily-mean tas)	279.59	279.59 (raw daily max)
tmin (day)	277.71 (= daily-mean tas)	275.84	275.84 (raw daily min)

tasmax > tas > tasmin now holds everywhere. Full unit suite passes (1849 tests, 36 new/updated), and the CMIP7 integration tests for all four variants pass WCRP compliance.

Note for reviewers: the ESM1-5 mapping got the same change. Its stream layout and naming convention are identical (it already uses fld_s03i230_max for sfcWindmax), but I could only verify values against ESM1.6 data — worth confirming ESM1-5 archives carry fld_s03i236_max/_min in their pe daily files.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 77.3%. Comparing base (b5d3b16) to head (eb74279).

Additional details and impacted files

@@          Coverage Diff          @@
##            main    #648   +/-   ##
=====================================
  Coverage   77.3%   77.3%           
=====================================
  Files         40      40           
  Lines       8524    8519    -5     
  Branches    1593    1587    -6     
=====================================
- Hits        6590    6589    -1     
+ Misses      1604    1602    -2     
+ Partials     330     328    -2     

Flag	Coverage Δ
unit	77.3% <100.0%> (+<0.1%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 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.