gh-153569: Add tokenizer source and cursor primitives

[pablogsal]

This adds the decoded source store, offset spans, and an offset-only cursor as the foundation for moving tokenizer state away from buffer pointers.

For example:

// Own decoded bytes by line.
_PyTok_SourceText source;
_PyTok_SourceInit(&source);

static const char text[] = "answer = 42\n";
_PyTok_Off line_start = _PyTok_SourceAppendLine(
    &source, text, sizeof(text) - 1, 0);
assert(line_start >= 0);

// Navigate by source offset.
_PyTok_Cursor cursor;
_PyTok_CursorInit(&cursor, &source);
assert(_PyTok_CursorSetOffset(&cursor, line_start) == 0);

_PyTok_Off name_start = cursor.pos;
while (_PyTok_CursorPeek(&cursor, 0) != ' ') {
    assert(_PyTok_CursorAdvance(&cursor) != EOF);
}

// Keep token text as a half-open span.
_PyTok_Span name = _PyTok_SpanFromBounds(name_start, cursor.pos);

// Borrow a temporary view.
Py_ssize_t name_len;
const char *name_text = _PyTok_SourceSpanView(&source, name, &name_len);
assert(name_text != NULL);
assert(name_len == 6 && memcmp(name_text, "answer", name_len) == 0);

_PyTok_SourceClear(&source);

This PR only adds these primitives and their direct tests. It does not migrate the existing readers, lexer and f-string state, pegen, or _tokenize; those changes can be built on this API in separate PRs.

• Issue: Base the tokenizer API on source offsets #153569