Fix evaluate for a piecewise linear SplineSpace

[jishnub]

evaluate for a SplineSpace{1} referred to an undefined f in its body
instead of its arguments:

function evaluate(c::AbstractVector{T}, s::SplineSpace{1,<:Real}, x::Real) where T
    p = domain(f).points
    c = f.coefficients
    ...

so evaluating such a Fun threw an UndefVarError. This isn't a dead code
path — integrate of a Fun in a HeavisideSpace returns a Fun in a
SplineSpace{1}:

julia> g = integrate(Fun(HeavisideSpace([-1.0,0.0,1.0]), [1.0,2.0]));

julia> g(0.5)
ERROR: UndefVarError: `f` not defined in `ApproxFunBase`

The loop bound is now set by the number of coefficients as well. A Fun may
have fewer coefficients than the dimension of its space, in which case the
interpolation would read past the end of the coefficient vector and throw a
BoundsError. This matches the corresponding method for a HeavisideSpace,
which evaluates to zero beyond the coefficients that are available.

The interpolation itself was correct, and is unchanged.

Tests check that the piecewise linear Fun interpolates its coefficients at
the nodes, that values(g) == g.(points(g)), that it evaluates to zero
outside the domain, that differentiate inverts the integrate above, and
that a truncated coefficient vector is handled.

Independent of #688.

🤖 Generated with Claude Code