Fix synthio oscillator wrapping one sample late

[bdbarnett]

Fixes #11266.

synth_note_into_buffer() in shared-module/synthio/__init__.c wraps the
DDS accumulator on

if (accum > lim) {
    accum = accum - lim + offset;
}

but lim is the exclusive end of the waveform loop -- the readable
samples are [offset, lim). Wrapping on > rather than >= lets the
accumulator land exactly on lim, and that iteration indexes
waveform[waveform_length]: one past the loop, and for the common case of a
note looping the whole table, one past the buffer itself. The ring-modulator
loop just below has the identical bug with ring_waveform.

Any note whose dds_rate divides lim hits the boundary on a schedule --
a table played at exactly one sample per output sample hits it every
waveform_length samples. When the loop covers the whole buffer, the read
is genuinely out of bounds, so the same script with the same events does
not render the same audio twice.

Repro

A table with a loop end short of the buffer end, so the extra read stays
in-bounds and lands on a marker value instead of undefined memory
(waveform_loop_end=256 of a 512-sample table, frequency set so the
oscillator advances exactly one table step per output sample):

Measured on a unix coverage build of main, four blocks of 256 samples
(1024 total), marker at index 256:

	before	after
marker hits	[255, 511, 767, 1023]	[]

Includes a regression test
(tests/circuitpython/synthio_oscillator_loop_end.py) built the same way;
it fails on current main and passes with the fix. The neighboring
biquad/filter tests that pass on unmodified main in the same environment
still pass.

[tannewt]

Thank you!