feat(gint): enable mixed-precision (fp32/fp64) support for GPU path

[dzzz2001]

Reminder

• Have you linked an issue with this pull request?
• Have you added adequate unit tests and/or case tests for your pull request?
• Have you noticed possible changes of behavior below or in the linked issue?
• Have you explained the changes of codes in core modules of ESolver, HSolver, ElecState, Hamilt, Operator or Psi? (ignore if not applicable)

Linked Issue

Follow-up to #7149 (CPU mixed-precision gint support)

Unit Tests and/or Case Tests for my changes

• Tested with 10 LCAO benchmark cases comparing gint_precision = double vs gint_precision = mix on GPU.

What's changed?

This PR extends the mixed-precision grid integration (gint_precision = mix/single) support from CPU-only to GPU. The key changes include:

GPU Kernel Templating

• phi_operator_gpu and phi_operator_kernel: Templated the PhiOperatorGpu class and associated CUDA kernels to support both float and double precision types.
• dgemm_vbatch and gemm_nn/tn_vbatch: Templated the batch GEMM operations to dispatch between float and double at compile time.

GPU Gint Functions

• gint_vl_gpu / gint_rho_gpu: Updated to use the templated GPU operators, dispatching precision based on the gint_precision parameter.
• gint_fvl_gpu, gint_tau_gpu, gint_vl_metagga_gpu, gint_vl_nspin4_gpu, gint_fvl_meta_gpu, gint_vl_metagga_nspin4_gpu: Propagated the precision template parameter through all GPU gint entry points.

Input Validation

• read_input_item_system.cpp: Removed the restriction that forced gint_precision back to double when running on GPU, allowing single and mix modes to work with GPU acceleration.

How Mixed-Precision Works on GPU

When gint_precision = mix:

1. Early SCF iterations use fp32 for gint_vl and gint_rho computations → faster kernel execution and reduced memory bandwidth.
2. Once charge density convergence (drho) approaches scf_thr, the GintPrecisionController switches to fp64 for final convergence accuracy.
3. Force and stress calculations always use fp64 regardless of precision setting.

Any changes of core modules? (ignore if not applicable)

• No changes to core modules (ESolver, HSolver, ElecState, Hamilt, Operator, Psi). All changes are confined to module_gint and input parameter validation.