[18.0][IMP] spreadsheet_oca: server-side pivot engine + scheduled data refresh

[dnplkndll]

Adds a server-side pivot engine to spreadsheet_oca, plus a scheduled refresh
that recomputes a spreadsheet's Odoo pivots and posts the tables to its Chatter.

The first two commits are shared by #100–#105 and will drop out of those PRs
once this one merges.

Non-mechanical points worth flagging

• pivot_data.py reimplements the JS PivotModel in Python. For row
  groupbys R and column groupbys C it queries every prefix pair — the
  "divisors" — which is what produces a pivot's subtotal rows, and costs
  (len(R)+1) x (len(C)+1) queries per pivot, matching the JS.
• Groups carry both ids and labels. rowValues stays id-based because
  grouping needs it; rowLabels is what anything human-facing renders, or a
  many2one group shows up as a bare database id.
• A failed subtotal is reported, not swallowed — the result is flagged
  partial with failedGroupBys, and an AccessError propagates rather than
  degrading into an empty group.
• One shared cron, with per-record identity. A single scheduled action
  refreshes whichever schedules are due, and each runs as its own Run As
  user (with_user), so a summary never contains rows that user cannot read.
  A static cron alone would run everything as the install user.
• read_group is deliberate. It returns __domain and formatted values
  that _read_group does not. The single call site is annotated as the one
  place to port for 19.0, where it becomes formatted_read_group.
• The [FIX] is an unrelated upstream bug: _compute_filename iterated
  self but read self.name, raising a singleton error on any multi-record
  set. It is split out so it can be reviewed and merged on its own.

AI-assisted (Claude Code); every change reviewed, tested, and owned by the author.