Added a dummy tag for ForwardDiff configs being constructed in Gridap Autodiff.jl to fix issue #805

[kishore-nori]

No description provided.

[amartinhuertas]

see #805 (comment)

[kishore-nori]

After making the last commit of changes the CI takes

Test Summary:  | Pass  Total
FESpaces (2/2) |  855    855
1699.394610 seconds (472.28 M allocations: 26.718 GiB, 0.88% gc time, 99.82% compilation time)

with a dummy tag function it took:

Test Summary:  | Pass  Total
FESpaces (2/2) |  855    855
516.246620 seconds (387.76 M allocations: 21.898 GiB, 1.95% gc time, 99.66% compilation time)

I think this is caused by the tag checking now (also having long tag's like we are ending up here may be has to do something with this, I am not sure, will have to look into this, haven't got much details from profiler)(sorry the tag check is not happening at all here, due to low-level API usage, the next or below might be the reason). Or it could be I am doing something horribly bad in the last commit (I don't feel so, as the changes have been very minimal, just adding a function to existing structs to extract the tag name while setting up ForwardDiff configs)

Update: Digging through the code, I also feel this could be because of type instability in inferring the return of type of _change_argument (the ForwardDiff tag is now being generated from the output of _change_argument, this might be propagating the type instability of pre-existing type-instability of _change_argument), may be better to type annotate the output of the _change_argument function as Function, I am not entirely sure if this can help.