Record attribute definitions that use xpath can break bib ingest

Record attribute definitions that use XPath to extract their values can break ingest. For example, consider the following definition:

-[ RECORD 1 ]+------------------------------------------------------
name | content_type
label | Content Type
description |
multi | t
filter | t
sorter | f
composite | f
tag |
sf_list |
joiner |
xpath | //marc:datafield[@tag='336']/marc:subfield[@code='a']
format | marcxml
start_pos |
string_len |
fixed_field |
phys_char_sf |

Attempting to ingest a bib that has a 336$a will fail with the following error:

CONTEXT: SQL statement "SELECT metabib.reingest_record_attributes(NEW.id, NULL, NEW.marc, TG_OP = 'INSERT' OR OLD.deleted)"
PL/pgSQL function biblio.indexing_ingest_or_delete() line 50 at PERFORM
ERROR: invalid XML content
DETAIL: line 1: AttValue: " or ' expected
{"<subfield code=\"a\">two-dimensional moving</subfield>"}
                 ^
line 1: attributes construct error
{"<subfield code=\"a\">two-dimensional moving</subfield>"}

Evergreen master