Speed up fixed-effect absorption

[matthieugomez]

• Avoid group permutations for intercept-plus-slope blocks using stable anchored weighted moments.
• Add adaptive bucketed CUDA gathers and align Metal gather planning with the same strategy.
• Cover weighted slopes, reversed interaction order, large offsets, and GPU parity.

[codecov]

Codecov Report

❌ Patch coverage is 48.24561% with 59 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (main@bdd9d6c). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
ext/CUDAExt.jl	0.00%	47 Missing ⚠️
ext/MetalExt.jl	0.00%	7 Missing ⚠️
src/AbsorptionPlan.jl	92.59%	4 Missing ⚠️
src/AbstractFixedEffectLinearMap.jl	83.33%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main      #85   +/-   ##
=======================================
  Coverage        ?   66.06%           
=======================================
  Files           ?       12           
  Lines           ?      940           
  Branches        ?        0           
=======================================
  Hits            ?      621           
  Misses          ?      319           
  Partials        ?        0           

☔ 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.