BeanWrapperFieldSetMapper ignores FieldSet#read* methods

Using Spring Batch version 2.2.2.
Want to use a FlatFileItemReader to parse a simple CSV file. Numbers inside CSV have commas as decimal separators.
Building the reader as follows:

@Bean
public FlatFileItemReader<MyTarget> reader() {
    DefaultFieldSetFactory fieldSetFactory = new DefaultFieldSetFactory();
    fieldSetFactory.setNumberFormat(NumberFormat.getInstance(Locale.getDefault()));

    return new FlatFileItemReaderBuilder<MyTarget>()
            .name("myReader")
            .resource(new FileSystemResource("a/filename"))
            .delimited()
            .delimiter(";")
            .fieldSetFactory(fieldSetFactory)
            .names(new String[] { "field1", "field2", "field3" })
            .targetType(MyTarget.class)
            .build();
}

Checked that fieldSetFactory has the right decimal separators.

This ItemReader works well with dot-separated decimal number, but fails to work with commas as decimal separators. Tested with fully custom FieldSet but DefaultFieldSet from DefaultFieldSetFactory is also affected.
The appropriate read*methods (readFloat for instance) are never called. Instead BeanWrapperFieldSetMapper converts the given input by using ConversionService and PropertyEditors. This might be a good idea in general but leads to ignorance of FieldSet's read* methods and thus makes it impossible to configure NumberFormats and DateFormats (among other things).

Start with BeanWrapperFieldSetMapper and trace how it maps values instead of calling FieldSet#read* methods; compare that path with DefaultFieldSet and the configured NumberFormat. Reproduce the locale-specific decimal case from the issue and add regression coverage showing the intended FieldSet conversions are honored.