EnglishRoberta.TokenizeToIds only populates the accumulatedIds if found in cache

[stephentoub]

If the data isn't found in the cache, nothing is added to the list.

machinelearning/src/Microsoft.ML.Tokenizers/Model/EnglishRoberta.cs

Lines 262 to 305 in 4635a86

	private int TokenizeToIds(string sequence, IList<int>? accumulatedIds)
	{
	if (_cache.TryGet(sequence, out IReadOnlyList<Token>? hit))
	{
	if (accumulatedIds is not null)
	{
	foreach (var t in hit)
	{
	accumulatedIds.Add(t.Id);
	}
	}
	return hit.Count;
	}
	Span<char> token = stackalloc char[100];
	Span<int> indexMapping = stackalloc int[100];
	if (sequence.Length > 100)
	{
	token = new char[sequence.Length].AsSpan();
	indexMapping = new int[sequence.Length].AsSpan();
	}
	int newTokenIndex = 0;
	for (int i = 0; i < sequence.Length; i++)
	{
	if (_byteToUnicode.TryGetValue(sequence[i], out var value))
	{
	token[newTokenIndex] = value;
	indexMapping[newTokenIndex] = i;
	newTokenIndex++;
	}
	}
	if (newTokenIndex == 0)
	{
	return 0;
	}
	IReadOnlyList<Token> result = EncodeToTokens(token.Slice(0, newTokenIndex), indexMapping);
	_cache.Set(sequence, result);
	return result.Count;
	}

[stephentoub]

cc: @tarekgh