trapezoidal accepts the zero-blend velocity boundary but returns NaN at the initial sample #924

Reproduction

import numpy as np
from roboticstoolbox import trapezoidal

trajectory = trapezoidal(0.0, 1.0, np.array([0.0, 0.5, 1.0]), V=1.0)
print(trajectory.q)
print(trajectory.qd)
print(trajectory.qdd)

On current main 879fb363580840e19ee70b6b175e8cd3543e2b38 this prints:

[nan 0.5 1. ]
[nan 1.  1. ]
[inf 0.  0. ]

The accepted lower boundary abs(V) == abs(qf-q0)/T gives tb == 0; TrapezoidalProfile.__init__ divides by tb, and evaluating the initial blend multiplies that infinite acceleration by zero. Reversing the move reproduces the same non-finite initial sample. Controls at V=1.5 and V=2.0 return finite arrays.

Expected: handle the accepted boundary with a defined finite profile, or reject it explicitly if zero-duration blends are unsupported, rather than returning NaN/inf.

Reproduced with the normal source-package import on Linux, Python 3.12.14, NumPy 2.3.5 and SciPy 1.17.0. No robot hardware was used. This is separate from the empty-phase plotting fix in #910; filed following the request there. Searches for trapezoidal with blend and NaN found no existing issue.