Implement aten::_standard_gamma frontend conversion, add statistical tests, and update op-table mappings

[saquibjawedbit]

Details:

• Implemented full support for aten::_standard_gamma(Tensor self, Generator? generator=None)
  • Rejects non-None generator inputs
  • Promotes concentration to f32 internally and converts back to original dtype at the end
• Added shape extraction and broadcasting logic to produce the expanded sampling shape used by the algorithm.
• Implemented the Marsaglia–Tsang acceptance–rejection gamma sampling algorithm using OpenVINO ops:
  • RandomNormal
  • RandomUniform
  • Sqrt, Log, Power
  • Multiply, Add, Subtract, Maximum
  • CumSum, ReduceSum, ReduceLogicalOr
  • Gather, Select, Convert
• Added constants for:
  • number of sampling trials
  • minimum uniform cutoff value
  • fallback path using gamma_candidates or last_candidate depending on acceptance mask
• Final node output uses ConvertLike to match the original input dtype.

Added Tests

• Created tests that wrap torch._standard_gamma in a minimal PyTorch module.
• Converted this module with ov.convert_model using example inputs.
• Compiled the resulting model using:

  ov.Core().compile_model(...)

Tickets:

• Closes [Good First Issue]: Support aten::_standard_gamma #29686