bugfix(jobStatusPage): stop layer jump on Z-hop + harden print-stats slots

[gmmcosta15]

Description

Select the type:

• Feature
• Bug fix
• Code refactor
• Documentation

Follow-up hardening for the jobStatusPage layer logic introduced in #208. Fixes a field-observed bug where the current-layer counter jumped whenever the toolhead raised Z (travel moves / Z-hop) to reach another region of the plate, and aligns the layer-source priority with Mainsail's getters.

• Reject transient Z rises: the current layer now commits an advance only once Z has settled (confirmed across two samples) and never regresses, so Z-hop and travel moves no longer bump the layer number.
• Respect reported totals: a Klipper-reported print_stats.info.total_layer (or current_file.layer_count) is no longer overwritten by the geometry estimate, matching Mainsail's getPrintMaxLayers priority.
• on_flowguard_update: renamed params that shadowed the str/dict builtins, added type hints, and applied each key independently so a partial delta no longer raises KeyError.
• handleCancel: replaced the disconnect+reconnect dance with a single UniqueConnection (guarded) connect.
• showEvent: now calls super().showEvent().
• Simplified toggle_thumbnail_expansion, merged the duplicate valid-state branch in _handle_print_state, and tidied the print-stats dispatch.

Motivation

The Z-based layer fallback read the instantaneous gcode_move.gcode_position[2], so any non-print Z motion (Z-hop, travel to a taller area) was interpreted as a layer change and the counter flickered. Since gcode_position is published on essentially every move, a two-sample "Z settled" filter reliably distinguishes a real layer advance from a transient hop while keeping the counter monotonic.
The layer-source priority was also corrected to match Mainsail so a reported total layer count is never clobbered by the geometry estimate.

Tests

New/updated cases: transient Z-hop does not bump the layer, layer never regresses, reported total_layer not overridden by estimate, flowguard partial vs full payload, plus the two-sample settle behavior.