Add the "powerLaw" spread link

[eliotmcintire]

Adds a third way to turn the fitted covariates into a spread probability, built from a percolation calibration. Under it a wet or dry year can no longer push every fire past the percolation threshold. It is off by default.

Affects: Nothing changes unless link = "powerLaw" is asked for; the default link gives the same numbers and the same cache keys as before.

Technical details

Before this, the year random effect was added on the logit of the final p (objFunSpread.R, objFunInner()), which lets a year take p past the percolation threshold. New exported powerLawLink() (helpers.R) is p = pc - (pc - floor) * exp(-x / gamma) with x the linear predictor, pc in the maxAsymptote slot (fitted), floor = lowerSpreadProb, gamma fixed (spreadPowerLawGamma = 1.9, not fitted because it is confounded with the coefficient scale), and p < 0 set to 0. logisticAll(link = "powerLaw", gamma) calls it, so spreadProbFromCovs(), .objfunSpreadFit(), spreadProbGates(), spreadFitValidationData() and runDEoptim() take link = "powerLaw" and powerLawGamma; with it the year effect is added to x. powerLawOmitArgs() leaves powerLawGamma out of runDEoptim()'s cache key and re-score arguments unless the link is "powerLaw". New constants: spreadFitPowerLawGammaTxt (ledger column), "powerLaw" in spreadLinkOptions, spreadPowerLawPcBounds (0.14 to 0.35, the bounds of pc; runaway simulations on five real ELF landscapes put the threshold at 0.258-0.288, so spreadProbCeiling 0.276 would bind). Combining it with climateRef stops.

Tests: new test-powerLaw.R (link values, "ceiling" unchanged and bit-identical, year effect constant on x and not on the logit, gates, validation, runDEoptim() arguments and keys); the toy fixture of test-climateOutside.R moved to helper-spreadLinkFixture.R to share it. test-powerLaw.R fails on development (no powerLawLink, unused argument powerLawGamma). Full suite: 2136 pass, 0 fail on this branch; 2094 pass, 0 fail on development. R CMD check --no-manual: 0 errors, 0 warnings, 0 notes. Scratch library only.

🤖 Generated with Claude Code

https://claude.ai/code/session_014G18Mik3fH4z5M5CyMSUS6

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 89.47%. Comparing base (e655536) to head (9edea3f).

Additional details and impacted files

@@               Coverage Diff               @@
##           development     #167      +/-   ##
===============================================
+ Coverage        89.44%   89.47%   +0.03%     
===============================================
  Files               51       51              
  Lines             4451     4466      +15     
===============================================
+ Hits              3981     3996      +15     
  Misses             470      470              

☔ 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.