This is another solution by twb using lynx with options to set
the encoding. Wanted to record here for future reference
as it uses the lynx -dump option I originally suggested.

 1 def strip_html():
 2     '''Parse global "message_string" variable as a MIME message.
 3        Look for text/html MIME objects,
 4        run lynx -dump on them,
 5        insert them as text/plain MIME objects.
 6        Return the new message (as a string).
 7 
 8        NB: this often results in a multipart/alternative branch with 
TWO text/plain leaves,
 9        but that actually seems to work out pretty well.'''
10 
11     message_object = email.message_from_string(message_string)
12 
13     # NB: this is the earliest point we can extract header fields.
14     syslog.syslog('SUBJECT {}'.format(message_object.get('Subject', 
'No Subject')))
15     syslog.syslog('MESSAGE-ID 
{}'.format(message_object.get('Message-ID', 'No Message-ID')))
16 
17     # NB: this walk traverses ALL nodes in a flattened tree,
18     # so we do not need to manually recurse on branch nodes.
19     # (unlike perl). --twb, Sep 2015
20     for part in message_object.walk():
21         if 'text/html' == part.get_content_type():
22             syslog.syslog('STRIPPED an html part')
23             # Pipe it through lynx to render as plain text.
24             #
25             # NOTE: postfix runs maxwell with C (not C.UTF-8) 
locale!
26             # This breaks non-ASCII for things like open(mode='wt') 
and check_output(universal_newlines).
27             # As a workaround leave everything as b'' bytes and pass 
encoding hints to lynx and set_payload.
28             #
29             # NOTE: It is very unintuitive, but
30             #           get_payload(decode=False) ⇒ u'…'
31             #           get_payload(decode=True)  ⇒ b'…'
32             #       This is because decode=True decodes only C-T-E: 
base64 (or quoted-printable);
33             #       the e.g. ISO-8859-1 to Unicode decoding happens 
later!
34             output = subprocess.check_output(
35                 ['lynx',
36                  '--dump',
37                  '--stdin',
38                  '--assume-charset', 
part.get_content_charset(failobj='UTF-8')],
39                 universal_newlines=False,  # KLUDGE — see above
40                 input=part.get_payload(decode=True),
41                 env={'LC_ALL': 'C.UTF-8'})
42             # Edit the part to be plain text.
43             # FIXME: instead of editing the old MIME object,
44             # *delete* it and create a new text/plain object.
45             # The only part we might want to keep is inline vs. 
attachment (disposition)?
46             # But how do I *do* that during a .walk()? --- it's not 
a foldr!
47             del part['Content-Type']
48             del part['Content-Transfer-Encoding']  # alloc #31444
49             part['Content-Type'] = 'text/plain'
50             part.set_payload(output, 'UTF-8')
51             # FIXME: should append '.txt' to filename=fred.html 
where present.
52             # I can't see how to change it without also overriding 
the disposition,
53             # which is undesirable. --twb, Sep 2015
54 
55     return message_object