Enhancement request: allow importer inheritance in CSVMeta

[Forrest-T]

The comment in importers/csvbase.py says "The CSV Importer class needs to inherit from beangulp.Importer". The implementation assumes direct inheritance, but I think we can support indirect inheritance fairly easily.

I ran into this limitation when I tried to model multiple accounts at the same bank with inheritance, which all share the same schema:

class MyBankCSVImporter(csvbase.Importer):
    '''Base class for the CSV schema used by my bank'''
    date = csvbase.Date("Date", "%m/%d/%Y")
    # CSV schema...

class MyBankCSVImporter_Checking(MyBankCSVImporter):
    '''Import data for checking account'''
    def identify(self, filename: str):
        return "checking_1234" in os.path.basename(filename)

class MyBankCSVImporter_Visa(MyBankCSVImporter):
    '''Import data for visa card ending in 5678'''
    def identify(self, filename: str):
        return "visa_5678" in os.path.basename(filename)

CSVReader's read(...) function uses self.columns.items() to dynamically generate a 'Row' type. For the derived classes like MyBankCSVImporter_Checking, self.columns is an empty dictionary because CSVMeta only considers data members added by the most-derived class.

My proposal would be that CSVMeta class should merge the columns added by the base class(es) in between dct and beangulp.Importer. What do you think?

 class CSVMeta(abc.ABCMeta):                                                
     """A metaclass that extracts column specifications from class members  
     and stores them in a columns dictionary keyed by the member name."""   
     def __new__(mcs, name, bases, dct):                                    
         columns = {}                                                       
         others = {}
+        for b in bases:
+            if issubclass(b, beangulp.Importer) and b != beangulp.Importer:
+                columns = b.columns | columns
         for key, value in dct.items():                                     
             if isinstance(value, Column):                                  
                 columns[key] = value                                       
                 continue                                                   
             others[key] = value                                            
         others['columns'] = columns                                        
         return super().__new__(mcs, name, bases, others)