- survey.py: previous code would output position-based and
  reference-based translations (e.g. mylist-0 and choice:label), so
  it now skips reference-based translations for select_types.
- survey_element.py: related to the above change, test_xform2json.py
  test_convert_toJSON_multi_language failed due to circular reference.
  Change traverses translations to remove offending objects.
- pyxform_test_case.py: added the test case number to the debug/fail
  output, to help when there are many XPath expressions.
- numerous test fixes to accommodate updated choice output.

- now covered by tests/xpath_helpers/choices.py

- The test was broken by changes to choices processing (XLSForm#603).
- The test conditions seem to relate only to the instance_xmlns feature.
- The new test asserts the same using the pyxformtestcase pattern.

- OK locally but failing on CI

- builder.py: add type hint, remove unnecessary dict
- xform2json.py:
  - _set_survey_name now handles multiple instances
  - _set_submission_info now sets itemset details from secondary
     instance, and attempts to parse choice filters
  - _set_translations now returns value to be set in __init__()
  - _get_text_from_translation now does media comparison as a list,
    since a dictkeysview is not a list
  - add _get_choices to include choices in internal xform representation
- xls2json.py
  - skip adding choice lists that aren't used in the xform to avoid
    extra unnecessary data. Maybe worth a warning or maybe not.
  - add secondary instance for table-list appearance, but reject it if
    it was being added with a choice filter (not supported).
- xform2json_test.py:
  - expand test_load_from_dump variables for easier debugging
  - add test cases for choice filters, field-list and table-list
  - these test cases can be refactored to PyxformTestCase at some stage
- xls2xform_tests.py: use equivalent test method instead of keyword
- update expected XML from inline choices to secondary itemset for the
  following (these can be refactored to XPath PyxformTestCase):
  - test_table_list.py
  - repeat_date_test.xml
  - xml_escaping.xml

- reasonably confident that these are the expected types

- section should import SurveyElement from where it is defined.
- A NotImplementedError is commonly used to indicate a base method that
  should be implemented in child classes, and it is used in
  SurveyElement for xml_control(). The Option and Tag classes don't
  seem to need this method, so adding an implementation that raises this
  error is a bit clearer.

- move choices logic off to new function add_choices_info_to_question()
  - exclude select with appearance=search() from itemset (per comment)
  - new func for readability and to clarify what info is used to
    determine choices structures
- survey.py: skip adding translations for choices, instead of just for
  choice_filter. Use type "Option" to detect choice elements which
  should be more reliable. Move get_xpath lookup under if condition so
  it's only called if needed.
- xls2json.py: fix lookup of select type "select one external" which
  has an alias "select_one_external". Docs use the underscore alias
  but the implementation is looking for the spaces alias.
- xlsform_test_case/base.py: simplify reporter function
- xlsform_spec_test.py: simplify repetitive tests by adding test func.
  Ideally these should be re-implemented as PyxformTestCase, but they
  cover a lot of scenarios so for now just simplifying the structure.
- test_expected_output/*.xml: update expected XML to match current
  behaviour, generally: 1) update body to refer to instance, 2) add new
  instances in model, 3) add new translations in itext, 4) remove old
  itext from translations.
- Misc minor fixes:
  - utils.py: fix deprecation warning for get_sheet_by_name.
  - replace in-line regex with pre-compiled regex, and replace match
    (only looks at start of string) with search (looks at whole string).
  - add more constants for commonly-used internal strings so that it's
    easier to find where they are used, added usages to xls2json.py.

- The right number of empty rows would be removed afterwards anyway, but
  for someone reading the code later it'd be better if that number was
  as stated.

- yes_or_no_question and xml_tests updated for new itemset output for
  selects. The default_survey_sheet test expected no default_form_name.
- add type hints to workbook_to_json and parse_file_to_json.

- test_translations.py:
  - remove xpath helpers now in xpath_helpers/choices.py
  - update test xpath assertions to match itemset behaviour
- choices.py:
  - improve xpath helpers to not require specification of position

- builder.py:
  - add or_other choice to survey-level choice dict
  - add type annotations
- constants.py: add new constant for "or specify other" internal suffix
- xls2json.py: add error for choice_filter + or_other. Main reason I
  suppose is that it's unknown how to filter the "other" choice so the
  user might never see it.
- xlsform_spec_test.py:
  - fix or_other expected XML to include "other" and match sorted
    element names.
  - move or_other test into this testcase since it is the same as others
    there already.
- builder_tests.py: update expected output to match or_other output.