feat: add show methods for spaces, tensors and BilinearOperator

[wardjm]

Closes #15.

Base.show was defined only for LinearOperator and BettiSpectrum; everything else printed as the default struct dump, which for a tensor splats the whole coefficient vector into the REPL.

Adds compact show methods whose formats mirror the upstream Python __repr__s:

FunctionSpace(dim=50)
FormSpace(degree=1, dim=24)
DirectSumSpace(spaces=[FunctionSpace, VectorFieldSpace], dim=74)
ScalarFunction(space=FunctionSpace(dim=50), shape=(50,), batch_shape=())
DirectSumElement(space=DirectSumSpace(spaces=[FunctionSpace, VectorFieldSpace], dim=74), shape=(74,), batch_shape=())
LinearOperator(domain=FunctionSpace(dim=50), codomain=FunctionSpace(dim=50), shape=(50, 50))
BilinearOperator(domain_a=VectorFieldSpace(dim=24), domain_b=VectorFieldSpace(dim=24), codomain=VectorFieldSpace(dim=24), component_shape=(24, 24, 24))

The generic space/tensor methods sit in base_tensor_space.jl and base_tensor.jl; FormSpace and DirectSumSpace override them to carry the degree and the summand list. Coefficients are never printed. LinearOperators existing show is reformatted to the domain=/codomain= spelling, which the un-skipped upstream assertion requires.

Tests

The three previously-skipped repr assertions in test/test_pyclasses.jl are un-skipped, plus a show testset in test_tensor_algebra.jl covering what upstream does not check (FormSpace degree, DirectSumSpace summand list). Full suite: 2441 passing, up from 2417.

Note

Several @assert messages interpolate a space ("$(a.space)"), so their error text now reads FunctionSpace(dim=50) rather than a struct dump. An improvement, but a change beyond the REPL.