fix: remove redundant cache load and avoid GSettings lookup on every tick

[glazari1]

Hey @corecoding, here's the follow-up with the two items from my comment.

Summary

• Extract _ensureTimeSeriesLoaded() to deduplicate the cache load + merge
  logic (now used on first hover and on disable — prevents overwriting
  persisted history when the user never opened the popout)
• Remove redundant loadTimeSeries() from _init() — the lazy-load
  handles it with proper merge
• Cache show-sensor-history-graph in a _recordTimeSeries flag instead
  of calling get_boolean() on every sensor tick (6 calls/sec with
  default settings)
• Keep _recordTimeSeries and _timeSeriesLoaded in sync when toggling
  the setting on/off

Changes

extension.js:

• New _ensureTimeSeriesLoaded() method (extracted from _showHistoryPopout)
• _showHistoryPopout() calls _ensureTimeSeriesLoaded() instead of inline merge
• disable() calls _ensureTimeSeriesLoaded() before saving
• Signal handler updates _recordTimeSeries and _timeSeriesLoaded on toggle
• Remove loadTimeSeries() from _init()

values.js:

• Add _recordTimeSeries flag initialized from settings in _init()
• _pushTimePoint() checks cached flag instead of get_boolean()

Test plan

• Enable graph, collect data, hover sensor — graph shows correctly
• Enable graph, collect data, disable extension without hovering — re-enable and check history is preserved
• Toggle graph off/on via settings, verify data recording stops/resumes
• Monitor with pidstat — no GSettings overhead on every tick

[corecoding]

I forgot to push a change before your last PR, now there is a minor conflict. My primary change is below - this gets rid of two for loops. By the way, I ended up submitting an update to Vitals with the History feature removed. Another person submitted a bug report. I would like to get more people verifying this PR fixes the issue. In all honesty, even with 200 instead of 3600 datapoints, I still feel like there will be a stutter, albeit smaller.

        const memSeries = this._values._timeSeries;
        const memFormat = this._values._timeSeriesFormat;

[glazari1]

Thanks for the update! I understand the decision to remove the feature in v74 while it's being validated.

For what it's worth, I've been running the graph-history branch locally (Ryzen + GNOME 46, Ubuntu 24.04) and it's been completely smooth, measured ~12% average CPU on gnome-shell with zero GC spikes over multiple 15s samples using pidstat (was 60-70% spikes every ~11s before the fix). The widget reuse in history.js seems to make a big difference alongside the downsample.

Let me know if you'd like me to help with anything else or if more testing data would be useful.