Fix upserting when Synapse column name contains space or special characters

[andrewelamb]

Problem:

upsert_rows misreports how many rows need to be updated when a Synapse
table or view has a column name that is not a valid Python identifier, such
as a name with a space (AMP AIM_SSc-Skin_Access Team) or a hyphen
(col-with-hyphen).

• Reported by Jessica Vera during QA testing of SYNPY-1912
  (against syn76143756, a table with several columns whose names contain
  spaces).
• _construct_partial_rows_for_upsert in
  synapseclient/models/mixins/table_components.py loops over the freshly
  queried rows with results.itertuples(index=False). itertuples builds a
  namedtuple from the column names, but a namedtuple field name must be a
  valid Python identifier. When a column name has a space or another
  special character, pandas silently swaps in a fallback name for that
  field (e.g. _5), so the real column name is no longer reachable via
  getattr/hasattr.
• The code used getattr(row, column) if hasattr(row, column) else None to
  read each cell, and if not hasattr(row, column): values_differ = True
  to decide whether a row changed. Since hasattr was always False for
  any column with a space or special character, every row was treated as
  changed for that column, on every call, even when nothing changed.
• This made every upsert against such a table report the full row count as
  "rows to update," regardless of how many rows actually changed.
• This bug pre-dates SYNPY-1912 (introduced in [GEN-2381] Pandas handling of nullable cells #1272, 2025-11-20). SYNPY-1912
  ([SYNPY-1912] Fix upsert_rows misreporting Table update responses #1446) fixed a separate, downstream issue in how update results were
  logged, but did not touch this code path.

Solution:

• Replaced attribute-style row access (getattr/hasattr on the
  itertuples result) with position-based access, which works regardless
  of what pandas named the namedtuple field.
• Built a column_positions map ({column_name: index}) from
  results.columns once per call, and used it to look up ROW_ETAG,
  ROW_ID, id, the primary key columns, and each compared column by
  position (row[column_positions[...]]) instead of by name.
• This is a minimal, targeted fix: it keeps the existing itertuples loop
  and comparison logic, and only changes how values are read out of each
  row.

Testing:

• Added
  test_construct_partial_rows_for_upsert_with_column_name_containing_special_characters
  to tests/unit/synapseclient/mixins/unit_test_table_components.py. It
  exercises columns named col with space and col-with-hyphen with
  identical values in the queried results and the upsert data, and asserts
  that no rows are reported as changed. This test failed before the fix
  (assert 2 == 0) and passes after it.
• Ran the full existing test file
  (tests/unit/synapseclient/mixins/unit_test_table_components.py): all 336
  tests pass, including the 13 pre-existing tests that call
  _construct_partial_rows_for_upsert directly, so no regressions.
• Verified against the real table (syn76143756) that its column names do
  in fact contain spaces and special characters (e.g.
  AMP AIM_SSc-Skin_Access Team), confirming this fix addresses the
  reported case.