Optimization.

Suggested Optimization: Lazy Per-Account Loading + Avoid Full Reload on .reload config

Currently, the module loads all unlocked appearances during the OnAfterConfigLoad hook, which means the entire custom_unlocked_appearances table is read every time .reload config is executed.
On large servers this can exceed 1.8 GB of data being loaded into memory at once.

Current code:
```
`void OnAfterConfigLoad(bool reload) override
{
sT->LoadConfig(reload);
if (sT->GetUseCollectionSystem())
{
LOG_INFO("module", "Loading transmog appearance collection cache....");
uint32 collectedAppearanceCount = 0;
QueryResult result = CharacterDatabase.Query("SELECT account_id, item_template_id FROM custom_unlocked_appearances");
if (result)
{
do
{
uint32 accountId = (*result)[0].Get();
uint32 itemId = (*result)[1].Get();
if (sT->AddCollectedAppearance(accountId, itemId))
{
collectedAppearanceCount++;
}
} while (result->NextRow());
}
LOG_INFO("module", "Loaded {} collected appearances into cache", collectedAppearanceCount);
}
}`
```

Do not load the full table inside OnAfterConfigLoad, since this hook runs every time .reload config is used.

Implement lazy loading per-account, for example:

load a player’s unlocked appearances only when they log in;

optionally free the cache on logout to reduce memory usage.

Keep OnAfterConfigLoad strictly for config reloads, not for large dataset initialization.

Benefits:

No heavy delay or memory spike when using .reload config.

Significantly reduced memory usage.

Cache size scales with active players, not with total table size.

Faster worldserver startup.

This would be a strong improvement for large servers with big transmog datasets.

@Nyeriah

Start with OnAfterConfigLoad and the custom_unlocked_appearances query shown in the issue, then trace where account appearance data is accessed during login and logout. Keep configuration reloads free of full-table loading; done means account data loads on demand, optional logout cleanup works, and the cache scales with active players.

Written by the indexing model from the issue text.