Pin demand aggregation keyword behavior (#126)

[PabloBotin]

Closes Sienna-Platform#126.

Context

Issue Sienna-Platform#126 reports that plot_demand/plot_demand! document an aggregate keyword that is not actually used, while PowerAnalytics expects a typed aggregation.

Investigating on main, the functional gap is already closed: _translate_demand_aggregate rewrites aggregate (string or type) into aggregation, and a directly-supplied aggregation passes through untouched. What was missing is that neither spelling was pinned by a test, and the docstrings still documented only the string form — which is how the mismatch went unnoticed in the first place.

So this is a test-and-docs change; src/ behavior is unchanged.

Changes

• test/test_demand_aggregation.jl (new) — unit tests over _translate_demand_aggregate: all three string mappings (with :aggregate removed and :aggregation set), the typed passthrough issue Add aggregation keyword argument to plot_demand and plot_demand! Sienna-Platform/PowerGraphics.jl#126 asked about, a type supplied under the legacy aggregate name, aggregate = nothing as a no-op, and the ArgumentError on an unknown string. A final assertion checks every mapped type falls inside the set PA.get_load_data aggregates over, so a PowerAnalytics signature change is caught rather than silently ignored.
• test/test_plot_creation.jl — end-to-end pin that aggregation = ACBus survives to get_load_data. Reuses the existing sys_with_ts, passes no save, so the expected-file assertion below it is unaffected.
• src/call_plots.jl — docstrings only. aggregation::Type is documented as the primary spelling with the four forms PowerAnalytics accepts, aggregate::String as the convenience alias, and both now note it only applies when plotting a System (PowerAnalytics silently ignores it for Results). Corrects PowerSystems.Bus to PowerSystems.ACBus.

Verification

• Full suite green: 52 passed / 52 total, exit 0, zero Error log events.
• Formatter clean.
• Regression-checked: deleting the translation line in _translate_demand_aggregate makes the new test file fail (1 failed, 4 errored), confirming the tests are load-bearing.

Known limitations

• The new end-to-end case asserts plot_length == 3, the same count as the pre-existing aggregate = "Bus" case; it proves the typed keyword survives the translation path but is not independently load-bearing for detecting a dropped keyword — the existing aggregate = "System" case (expects 1) carries that signal.
• "PowerLoad" is covered by the type-set assertion but not end to end.
• The docs build could not be verified locally (DocumenterInterLinks inventory fetches fail on a self-signed cert in this environment; pristine HEAD fails identically). The @extref targets added are all already used elsewhere in the same file, but CI should confirm.