Use the real constants module.

This is progressing refactor of the calibre code to make it more
readable, and transform it to something more coherent.

In this patch, there are changes regarding imports for some modules,
instead of polluting namespace of each module with some other modules
symbols, which often were imported from other modules. Yuck.

ebook_converter/ebooks/mobi/reader/mobi8.py

@@ -8,6 +8,7 @@ import uuid
 
 from lxml import etree
 
+from ebook_converter import constants as const
 from ebook_converter.ebooks.mobi.reader.headers import NULL_INDEX
 from ebook_converter.ebooks.mobi.reader.index import read_index
 from ebook_converter.ebooks.mobi.reader.ncx import read_ncx, build_toc
@@ -17,7 +18,7 @@ from ebook_converter.ebooks.metadata.opf2 import Guide, OPFCreator
 from ebook_converter.ebooks.metadata.toc import TOC
 from ebook_converter.ebooks.mobi.utils import read_font_record
 from ebook_converter.ebooks.oeb.parse_utils import parse_html
-from ebook_converter.ebooks.oeb.base import XPath, XHTML, xml2text
+from ebook_converter.ebooks.oeb.base import XPath, xml2text
 from ebook_converter.polyglot.builtins import as_unicode
 
 
@@ -553,8 +554,8 @@ class Mobi8Reader(object):
         seen = set()
         links = []
         for elem in root.iterdescendants(etree.Element):
-            if reached and elem.tag == XHTML('a') and elem.get('href',
-                    False):
+            if reached and elem.tag == const.XHTML_A and elem.get('href',
+                                                                  False):
                 href = elem.get('href')
                 href, frag = urllib.parse.urldefrag(href)
                 href = base_href + '/' + href