Bioschemas event dates are not ISO 8601 and ignore the event's timezone #1391

(Claude found this one, sorry)

It seems to affect .json, .jsonld and .json_api

Problem

The Bioschemas JSON-LD of events (/events/<id>.jsonld and the copy embedded in the event page) gives
startDate and endDate in Ruby's default Time#to_s format, and labels the event's local wall-clock time as UTC:

"startDate": "2015-11-23 10:16:33 UTC"

There are two separate problems:

1. Not ISO 8601. schema.org startDate/endDate expect a Date or DateTime
   in ISO 8601 format, e.g. 2015-11-23T10:16:33-05:00. Consumers that parse the value strictly (search engines'
   structured data, JSON-LD/RDF tooling with xsd:dateTime) reject or misread it.
2. Wrong instant for events outside UTC. TeSS stores an event's start/end as local wall-clock time in a
   UTC datetime column, with the zone in the separate timezone field (see Event#start_local/#start_utc).
   The generator emits the raw column, so the local time is presented as UTC.

Example (test fixtures)

Cause

• lib/bioschemas/event_generator.rb and lib/bioschemas/course_instance_generator.rb map startDate/endDate
  straight to the start/end columns:

  property :startDate, :start
  property :endDate, :end

• Bioschemas::Generator#to_json uses JSON.pretty_generate, which serialises an ActiveSupport::TimeWithZone
  with to_s ("2015-11-23 10:16:33 UTC") rather than with ActiveSupport's ISO 8601 as_json.

• test/integration/bioschemas_test.rb asserts the current format (e.g. '2021-09-20 23:15:00 UTC'), so the
  tests encode the bug.

Materials are not affected: their dateCreated/dateModified/datePublished are date columns and already
serialise as ISO 8601, e.g. 2020-01-01.

Suggested fix

1. Emit local time with its offset in ISO 8601

property :startDate, ->(event) { event.start_local&.iso8601 }
property :endDate, ->(event) { event.end_local&.iso8601 }

in both generators, and update test/integration/bioschemas_test.rb. For the fixtures this gives:

2. Don't claim UTC when the timezone is unknown

When an event has no timezone, start_local falls back to the stored value, giving 2020-01-01T09:00:00+00:00.
That still asserts UTC. It would be more accurate to emit the local time without an offset
(2020-01-01T09:00:00), which ISO 8601 and schema.org allow and which means "local time, zone not known".

3. Fill in a missing timezone from the event's location

Many events have no timezone (see the TODO in Event#set_default_times, "Set timezone for online events. Where to get
it from, though?"), but in-person events are already geocoded to latitude/longitude by the Nominatim
geocoding worker. The timezone could be derived there, only when timezone is blank, so a timezone given by the
user or the ingestion source is never overwritten. Options, tried against the TeSS fixtures and a few known
places:

A reasonable combination: point-in-polygon from the coordinates when they are known, otherwise the country when it
has a single zone, otherwise leave timezone blank (and emit times without an offset, as in 2). Note that
country is free text (fixtures include USA and üK), so a country lookup needs to tolerate names as well as
codes; the countries gem already in the bundle helps.

Online events have no location to derive a timezone from; they still rely on the user or the ingestion source.

Also affected

The Detached RO-Crate export #1390 reuses the Bioschemas properties, so it inherits this and is fixed by the same change.