fix(calendar): stop rebuilding a VTODO from the partial update dict on error

[thejdubb02]

Fixes #1252.

_merge_ical_todo_properties ended in a catch-all that returned
_create_ical_todo(todo_data, todo_uid). On the update path todo_data holds only the
fields the caller passed, so any exception replaced the stored todo with one rebuilt from
that partial dict — dropping summary, due date, dtstart, the completed timestamp,
categories, RRULE, alarms and every custom property the caller did not re-send — and
reported success.

This is the same defect removed from the event merge in #544. _merge_ical_properties
raises instead and its docstring records why; the todo path never got the same treatment.
It was already half-fixed — except ValueError: raise was added with a comment pointing
at that precedent — but every other exception still fell through to the rebuild.

todo_uid goes with it, for the reason event_uid did: the parameter existed only to
feed the fallback, and the UID carries through from the stored iCal as a preserved
property.

Reviewing

Most of the diff is the dedent from dropping the try:. ?w=1 shows the real change: 35 lines.

Test

test_a_non_value_error_does_not_silently_rebuild_the_todo passes a malformed reminder,
which raises AttributeError — precisely what the catch-all swallowed. It fails on
master and passes with this change. No mocking; the input is real.

On propagating instead of catching

One reviewer raised whether removing a catch-all risks unhandled exceptions reaching the
client. Worth stating what I checked: nc_calendar_update_todo catches only
DavPreconditionFailed and lets everything else propagate, and the event path has
behaved this way since #544. This change introduces no exception path that is not already
live for events. The alternative traded that for guaranteed silent data loss.

Provenance

Written with AI assistance, then verified rather than trusted:

• The full unit suite runs clean — 56 passed, and the 13 collection errors are
  pre-existing on untouched master (missing optional test deps in my environment), not
  introduced here.
• ruff check reports one fewer error than master; ruff format --check is clean.
• The new test was confirmed to fail without the source change and pass with it.
• The patch was reviewed by three independent models plus a human before submission, and
  the one dissent is answered above.
• Running against my own Nextcloud instance, where this code path is in daily use.

Happy to split the whitespace-only dedent into its own commit if that reads better.

[CLAassistant]

All committers have signed the CLA.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud