Simplify and speedup shape_gradient_and_value for embedded

[KnutAM]

Noticed this when reviewing #938

ip = Lagrange{RefQuadrilateral,1}()^3; ξ = rand(Vec{2});
@btime Ferrite.shape_gradient_and_value($ip, $ξ, 1);
# PR:     15.030 ns (0 allocations: 0 bytes)
# master: 26.459 ns (1 allocation: 64 bytes)

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 93.84%. Comparing base (c956e70) to head (6e2b4a8).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #947      +/-   ##
==========================================
- Coverage   93.84%   93.84%   -0.01%     
==========================================
  Files          36       36              
  Lines        5317     5310       -7     
==========================================
- Hits         4990     4983       -7     
  Misses        327      327              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[termi-official]

Since this is already done I am okay with merging this, but shouldn't we rather try to finalize Ferrite-FEM/Tensors.jl#188 (comment) ?

[fredrikekre]

Need to re-resolve the doc manifest to include ForwardDiff as a dependency.

[KnutAM]

but shouldn't we rather try to Tensors.jl#188

I would love to, but need quite some input on Ferrite-FEM/Tensors.jl#211 before doing a managable implementation of Ferrite-FEM/Tensors.jl#188 (and that will probably be after Ferrite 1.0)

I simply did this to check my understanding of the code, since @lijas re-uses this code in #938. So perhaps not clear from the PR, but since I then had this code the main merit is to simplify for others to see what is happening here, instead of relying on Tensor internals. (On the other hand, this code was probably beneficial for generalizing to the 3rd order case, but will look at that now if a similiar simplification can be suggested in 938).