Explain torch compile error messages, improve PTQ vs QAT doc

[andrei-stoian-zama]

Closes https://github.com/zama-ai/concrete-ml-internal/issues/4497
Closes https://github.com/zama-ai/concrete-ml-internal/issues/4387
Closes https://github.com/zama-ai/concrete-ml-internal/issues/4487
Closes https://github.com/zama-ai/concrete-ml-internal/issues/4498

[RomanBredehoft]

thanks ! But I don't think it entirely closes https://github.com/zama-ai/concrete-ml-internal/issues/4414 as this issue is more about the error message itself. But yes great to add it in the doc

[RomanBredehoft]

also might be worth to consider https://github.com/zama-ai/concrete-ml-internal/issues/4498 in this PR as well, if possible 🙏

[yuxizama]

I suggest the following format for users to easily find solutions when they encounter some errors. Wdyt?

Common Compilation erros

Bit width limits

Error message: This is the error message
Cause: This error can occur when rounding_threshold_bits is not used and accumulated intermediate values in the computation exceed 16-bits. TLU input maximum bit-width is exceeded
Possible solutions:

• Reduce quantization n_bits. However, this may reduce accuracy. When quantization n_bits must be below 6, it is best to use Quantization Aware Training.
• Use rounding_threshold_bits. This feature is described here. It is recommended to use the fhe.Exactness.APPROXIMATE setting, and set the rounding bits to 1 or 2 bits higher than the quantization n_bits
• Use pruning

Crypto-parameters

Error message: This is the error message
Cause: This is the cause
Possible sulutions:

• Solution 1
• Solution 2

Quantization

Error message: This is the error message
Cause: This is the cause
Possible sulutions:

• Solution 1
• Solution 2

[github-actions]

Coverage passed ✅

Coverage details

---------- coverage: platform linux, python 3.8.18-final-0 -----------
Name    Stmts   Miss  Cover   Missing
-------------------------------------
TOTAL    7885      0   100%

60 files skipped due to complete coverage.