[wavebyrd]

Since #200 (numpy rewrite), ftr and the RVGI columns raise on data with fewer rows than their window:

df = StockDataFrame.retype(ohlcv.head(3))  # any OHLCV frame with 3 rows
df["ftr"]   # ValueError: window shape cannot be larger than input array shape
df["rvgi"]  # ValueError: could not broadcast input array from shape (2,) into shape (0,)
StockDataFrame.sym_wma4(pd.Series([4, 2, 2]))  # same broadcast error

• _ftr passes the prices to np.lib.stride_tricks.sliding_window_view without checking that there are at least window of them.
• _sym_wma4 assigns np.convolve(..., mode="valid") to out[3:]. With fewer than 4 values, numpy swaps the arguments and returns 5 - n values, which don't fit.

Before #200 both returned zeros in this case. The other windowed helpers (_mad, linear_wma, linear_reg, _inertia) already return early on short input, so this adds a guard to these two. Results for data longer than the window are unchanged.

Tests: added test_ftr_shorter_than_window, test_rvgi_shorter_than_window and test_sym_wma4_shorter_than_window to test.py. They fail with ValueError on master and pass with the fix. The full suite passes (135), and flake8 stockstats.py test.py is clean.