Fix concurrent LiteDB access crash with lock() serialization

[Copilot]

A single LiteDatabase instance was shared across all LiteDbDocumentCollection wrappers with no synchronization, causing InvalidOperationException: Collection was modified; enumeration operation may not execute when concurrent reads/writes triggered mid-enumeration deserialization in LiteDB's BsonMapper.

Changes

LiteDbDocumentDatabase

• Adds a shared object _lock = new() field passed to every LiteDbDocumentCollection wrapper
• No disposal needed — plain lock objects are GC-managed

LiteDbDocumentCollection

• Accepts a shared object syncLock in the constructor
• Every method touching _collection (Insert, Update, Upsert, Delete, DeleteAll, FindById, Exists, Find, FindAll, Count) wraps the LiteDB call with a lock statement
• UpsertAsync holds the lock across the full exists-check + insert/update sequence — eliminating the TOCTOU race
• Methods are no longer async since all LiteDB operations are synchronous; they return Task.FromResult(...) directly

lock (_lock)
{
    var exists = _collection.Exists(Query.EQ("_id", new BsonValue(document.Id)));
    if (exists)
        return Task.FromResult(Result.Success(_collection.Update(document)));
    else
        return Task.FromResult(Result.Success(_collection.Insert(document) != null));
}