[ConvertParallelToGPU2] use dynamic shared memory when a kernel needs more than 48KiB

[xys-syx]

The TODO in getMaxSharedMem says a block can use more than 48KiB of shared memory, but only as dynamic shared memory:

Enzyme-JAX/src/enzyme_ad/jax/Passes/ConvertParallelToGPU.cpp

Line 2591 in 44f289b

// TODO we can use more than 48KB (below) but it is only

This PR makes convert-parallel-to-gpu2 use dynamic shared memory when a kernel needs more than 48KiB:

• the arrays are placed together in one dynamic shared memory region
• every launch of the kernel passes the size of that region
• right before each launch, the host calls cudaFuncSetAttribute(kernel, cudaFuncAttributeMaxDynamicSharedMemorySize, size). [CUDA requires this call before a launch can use more than 48KiB, and appling the setting only to the current device, so it is made before every launch

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 29.62%. Comparing base (9409410) to head (74d0485).
⚠️ Report is 30 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #3411   +/-   ##
=======================================
  Coverage   29.61%   29.62%           
=======================================
  Files         240      240           
  Lines       48460    48474   +14     
=======================================
+ Hits        14351    14359    +8     
- Misses      34109    34115    +6     

☔ 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.