Issue #91 — LEDSpicerUI layout bugs — changes made

[meduzapat]

Task-Url: #91

Bug fixes:

• Random colors list no longer carries over stale entries between project loads (boxRandomColors is now wiped unconditionally before repopulating).
• Settings > General tab: Switch/Label order swapped to match the rest of the app (Label first, then Switch), and labels given expand=True so all switches align flush-right within their row.
• Removed 3 dead .clear() calls in readConfigFile (redundant with load()'s own internal clear).
• Removed a stale, contentless TODO comment in DialogLinkEditor.cpp.
• Re-selecting the already-open project now offers a real revert (discard
  unsaved changes + reload from disk) instead of just an info message, when the project is dirty.

Save-time "All" group optimization:

• ConfigFile::save() now omits the "All" group from the saved file when it exactly matches the default device order (position-for-position), since the daemon auto-builds it identically in that case.
• Strip elements are expanded into their children (not the unlinkable parent) when building that comparison, matching daemon behavior.
• xmlSection() gained an allowEmpty parameter so <layout>'s defaultProfile/defaultProject attributes still get written even when the groups body ends up empty after "All" is omitted.

Code quality / review follow-ups:

• Moved ProjectFile.hpp / Storage/Element.hpp includes into ConfigFile.hpp (header carries the deps, not the .cpp).
• Multi-line comments in ConfigFile.cpp switched from stacked // to /* */.
• Extracted the strip-expansion and order-comparison logic out of save() into two named, testable static methods: expandDeviceElements() and matchesDeviceOrder().

Tests:

• Added 7 unit tests covering both extracted helpers: plain element order, strip-parent exclusion (children only, correct position), mixed plain+strip sequencing, and matching cases for matchesDeviceOrder (identical, reordered, size mismatch, strip children reordered).
• Added tests/mocks/MockDevice.hpp (reusable minimal Parent stand-in for a device), matching the existing MockData/MockBasicData conventions.
• Verified with a deliberate negative control (temporarily broke matchesDeviceOrder) to confirm the new tests actually catch the regression they're meant to catch, not passing vacuously.

Investigated, found not to be bugs:

• "Default profile not selected after loading" — confirmed to be expected behavior in Local Mode, where projects share one physical ledspicer.conf file (not a defect).
• Cascade delete on Element removal (dependency system correctly purges stale Group links, including in "All", when devices are wiped) — verified correct, no dangling pointers.