Fix frame labels in chapter 2 notebook - #50

## Summary

`trplot()` and `trplot2()` both ended with a bare `**kwargs` that was
never read anywhere in their bodies, so a typo'd keyword (e.g.
`framelabel=` instead of `frame=`) was silently accepted and ignored
instead of raising — it drew the frame with no error and no label at
all. This surfaced downstream in
[petercorke/RVC3-python#50](petercorke/RVC3-python#50),
where a notebook had exactly this typo for years with no test ever
catching it.

- Removed the dead `**kwargs` sink from both functions so an
unrecognized keyword now raises `TypeError`, same as any normal Python
call.
- `trplot`'s `anaglyph` branch only forwarded a small fixed subset of
parameters to each eye's recursive call, silently dropping `textcolor`,
`labels`, `originsize`, `origincolor`, `axislabel`, `axissubscript`,
`width` and `projection`. It also never returned `ax` and never actually
called `plt.show(block=...)`, so anaglyph mode ignored `block` entirely.
Fixed to forward the full applicable parameter set, return `ax`, and
call `plt.show()` once after both eyes are drawn.
- `trplot`'s "T is an iterable of transforms" branch forwarded every
parameter except `axissubscript`, and also never returned `ax`. Fixed.
- `tranimate`/`tranimate2` were splatting the *same* unfiltered kwargs
dict into `Animate()`/`Animate2()`'s constructor, the drawing call, and
`run()`'s animation-control call, relying on each one's own dead
`**kwargs` sink to drop what it didn't need. Now that `trplot`/`trplot2`
validate strictly, `run()`-only parameters (`movie`, `repeat`,
`interval`, `nframes`, `wait`) are split off before the drawing call.
- Fixed a long-standing typo in `test_pose2d.py::test_graphics`
(`T0=T2`, which was never a real parameter — `animate()`'s documented
one is `start=`) that this same silent-swallow bug had been quietly
masking since the test was written.

## New test coverage

Added `tests/base/test_animate.py`. While chasing the
`tranimate`/`tranimate2` regression above, I found that constructing a
`FuncAnimation` under a non-interactive backend (the Agg backend this
repo's whole test suite runs under) doesn't actually run its per-frame
`update()` callback — confirmed directly by checking `Animate2`'s
internal frame counter, which never advances past its initial value
unless something drives the animation. So the existing
`animate()`/`tranimate()` tests only ever proved construction didn't
raise, not that interpolation/drawing works.

The new tests force every frame through the real callback via
`FuncAnimation.save()` with `PillowWriter` (pure Python, no external
`ffmpeg` dependency) and assert `trinterp`/`trinterp2` actually ran
across a real `s: 0→1` sweep. This is what caught the
`tranimate`/`tranimate2` kwargs-splitting bug above in the first place.

## Test plan

- [x] Full test suite passes: `353 passed, 3 skipped` (skips are
pre-existing, display-dependent)
- [x] Fail/fix/pass verified for every behavioral change (confirmed each
new/modified test fails against the pre-fix code for the intended
reason, and passes after)
- [x] `SE3`/`SE2` `.plot()`/`.animate()` wrapper methods (which forward
`*args, **kwargs` straight through) verified to propagate the stricter
validation correctly

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>