[tmva][sofie] Add optimization modes in SOFIE

[sanjibansg]

This PR adds two separate optimization modes in SOFIE so as to control the level of optimization the user prefers. As of now, we have OptimizationLevel:kBasic and OptimizationLevel:kExtended. By default, OptimizationLevel:kExtended is enabled which can be changed by:

using namespace TMVA::Experimental;
SOFIE::RModelParser_ONNX parser;
SOFIE::RModel model = parser.Parse("model.onnx");
model.SetOptimizationLevel(SOFIE::OptimizationLevel::kBasic);
model.Generate();

Optimization levels with their corresponding flags:

• kBasic: Operator fusion (for example: fusing ReLU with a preceeding GEMM)
• kExtended: Reuse memory for intermediate tensors

[github-actions]

Test Results

    17 files      17 suites   4d 12h 53m 9s ⏱️
 2 738 tests  2 735 ✅ 0 💤 3 ❌
44 943 runs  44 940 ✅ 0 💤 3 ❌

For more details on these failures, see this check.

Results for commit bd16854.

♻️ This comment has been updated with latest results.

[guitargeek]

Thank you very much for this PR!

I have validated that this works now for our use case of doing reverse mode AD with Clad on the emitted code.

Maybe consider squashing the commits on merge, since the first commit is not a state where the tests would pass and the second commit is a fixup to the first commit (let's try to make the tests pass at every commit).

[lmoneta]

LGTM!