Args:
    start_date: Inclusive start of the date range, ISO format YYYY-MM-DD.
    end_date: Inclusive end of the date range, ISO format YYYY-MM-DD.
    status_category: Optional filter slug. Only "upcoming" is confirmed to work;
        other values frequently 400 -- if one fails, drop it and retry without it
        rather than guessing an alternate spelling.
    location_category: Optional filter slug, e.g. "in-person", "virtual", "hybrid".
    product_category: Optional filter slug, e.g. "app-engine". The full product
        list is long (from the UI's Product filter) -- pass whatever slug the
        caller gives, or slugify a plain label (lowercase, spaces to hyphens).
    type_category: Optional filter slug, e.g. "webinar", "workshop", "academy",
        "office-hours", "360-exchange".
    include_full_description: If true (default), fetch each matched event's own
        detail page to get the full description instead of the listing's
        "..."-truncated blurb. Costs one extra HTTP request per matched event --
        set to false for date ranges wide enough that this would be slow (a few
        dozen events is quick; a few hundred will take a while, sequentially).
    save_csv_path: If given, also write the results to a CSV file at this path
        (columns: category, title, description, date_text, parsed_date, location,
        url). Embedded line breaks in text fields are rendered as literal `<br>`
        rather than real newlines, so each event stays on one CSV row.

Returns a dict with `events` (list of event dicts) and `meta` (counts: how many
events matched, how many full descriptions were/weren't fetched, etc. -- surface
anything notable from this to the user rather than silently discarding it).