Speed up fixed-effect scatter multiplies

[matthieugomez]

Summary

• add a six-argument scatter path that handles the BLAS-style beta argument inside the scatter kernel
• fuse beta scaling with the first forward fixed-effect multiply on CPU, CUDA, and Metal
• specialize beta == 0 and beta == 1 cases, and make CPU solver/map buffers more concrete
• bump version to 3.3.1

Tests

• JULIA_NUM_THREADS=8 julia --project=. -e 'using Pkg; Pkg.test()'

[codecov]

Codecov Report

❌ Patch coverage is 23.88060% with 51 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (main@84b29c4). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
ext/CUDAExt.jl	0.00%	24 Missing ⚠️
ext/MetalExt.jl	0.00%	18 Missing ⚠️
src/AbstractFixedEffectLinearMap.jl	58.33%	5 Missing ⚠️
src/SolverCPU.jl	69.23%	4 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main      #82   +/-   ##
=======================================
  Coverage        ?   51.57%           
=======================================
  Files           ?       11           
  Lines           ?      729           
  Branches        ?        0           
=======================================
  Hits            ?      376           
  Misses          ?      353           
  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.