[FIX] deltatech_pos_stock: stock badge frozen on stale value in open sessions

[dhongu]

Problem

The stock badge on the POS product card (deltatech_pos_stock) could stay frozen at a stale
quantity for days in an already open session, even after real sales depleted the stock.

Root cause: qty_available on product.template is a non-stored computed field. A sale writes
on stock.quant, not on product.template, so the template's write_date never changes. The
POS incremental sync (pos.load.mixin) filters strictly on write_date > last_server_date, so
an already open session never gets the recalculated quantity re-sent - the badge stays at
whatever value was cached when the session's data was first loaded.

Reproduced and confirmed on a production instance (Damira): the badge showed a quantity that
matched real on-hand stock from several days earlier, not the current value, even though the
POS session had been opened fresh that same day.

Fix

stock.quant now pushes a live STOCK_SYNCHRONISATION bus notification - reusing the same
pos.bus.mixin channel core already uses for pos.config.notify_synchronisation (POS device
sync) - whenever a product's on-hand quantity changes, to every open POS session with
display_stock enabled. The frontend subscribes to this channel and merges the fresh
product.template data straight into the in-memory model, independent of the write_date-based
sync cycle.

Only quantity changes trigger a notification (not reserved_quantity), since the badge shows
on-hand stock (qty_available), not the forecasted/reserved quantity.

Test plan

• Fresh install on a clean database (deltatech_pos_stock + demo data): existing badge test
  passes unchanged.
• New test test_stock_badge_updates_live_without_reload: opens a POS session, confirms the
  initial badge value, then changes the product's stock quantity via RPC (simulating a sale
  from another till/channel) while the session stays open, and asserts the badge updates
  to the new value without any page reload.
• Sanity-checked the new test actually detects the regression: temporarily disabled the
  notification call and confirmed the test fails (badge stays at the old value); restored
  the fix and both tests pass again.
• pre-commit (ruff, eslint, pylint-odoo, prettier) passes clean.

Co-Authored-By: Claude Sonnet 5 noreply@anthropic.com