PTX: add PTXRSqrtFastPass to fold afn 1/sqrt(x) to nvvm.rsqrt.approx

[maleadt]

Pattern-match fdiv afn 1.0, sqrt afn(x) and emit nvvm.rsqrt.approx.{f,d} directly. Runs before PTXFDivFastPass/PTXFSqrtFastPass so the pattern isn't destroyed before instruction selection sees it.

x-ref JuliaGPU/CUDA.jl#3149 (comment), cc @vchuravy
Continuing the effort from #800, #804, #805

[codecov]

Codecov Report

❌ Patch coverage is 98.03922% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 76.23%. Comparing base (f7d7418) to head (65b5b91).
⚠️ Report is 2 commits behind head on main.

Files with missing lines	Patch %	Lines
src/ptx.jl	98.03%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #807      +/-   ##
==========================================
+ Coverage   75.95%   76.23%   +0.27%     
==========================================
  Files          25       25              
  Lines        4026     4077      +51     
==========================================
+ Hits         3058     3108      +50     
- Misses        968      969       +1     

☔ View full report in Codecov by Sentry.
📢 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.