Show a warning if special fasta headers format is violated

[kim-fehl]

In a large dataset of automatically downloaded sequences there can be names including "|" symbol.
I concatenate class and train/test labels also automatically.
So, when I try to analyze this file, there are uninformative error messages like:

• ValueError: could not convert string to float: 'P42577.2'
• ValueError: invalid literal for int() with base 10: '6LPD'

which are caused by incorrect fasta headers:

• P42577.2_sp|P42577.2|FRIS_LYMST|0|training
• 6LPD_pdb|6LPD|F|1|training

A simple check when importing the file could show a warning to the user.

[Superzchen]

Thank you for your suggestion. We will deal with this problem as soon as possible.

[li6in9muyou]

I added a helpful error message for ill-formed headers in PR #3.