Use identity instead of hash for caching subfields in collect_subfields

[ktosiek]

I've noticed computing __hash__ on each cache access is quite expensive, and not what the JS implementation does - they are caching subfield nodes based on argument identity.

This PR is based on #55 for the benchmarks, but it can work separately.

Results of pytest --enable-benchmark -k benchmark before and after the change:

------------------------------------------------------ benchmark: 1 tests ------------------------------------------------------
Name (time in s)                               Min     Max    Mean  StdDev  Median     IQR  Outliers     OPS  Rounds  Iterations
--------------------------------------------------------------------------------------------------------------------------------
benchmark_executing_introspection_query     1.1361  1.1782  1.1484  0.0170  1.1427  0.0152       1;1  0.8707       5           1
--------------------------------------------------------------------------------------------------------------------------------

---------------------------------------------------------- benchmark: 1 tests ----------------------------------------------------------
Name (time in ms)                                Min       Max      Mean  StdDev    Median     IQR  Outliers     OPS  Rounds  Iterations
----------------------------------------------------------------------------------------------------------------------------------------
benchmark_executing_introspection_query     774.3004  789.3146  779.2556  6.0870  778.8992  7.2825       1;0  1.2833       5           1
----------------------------------------------------------------------------------------------------------------------------------------

this change also shaved 1/3 of the time from the benchmark mentioned in #54.

[Cito]

Again thanks a lot for this contribution which really makes a big performance difference.

This issue is actually a regression after #45 where the Node class got a correct, but slow, hash method. So using id() here is better even if it may result in fewer cache hits (when field nodes are equal, but not identical as objects).

Can you remove the benchmarking stuff from this PR, so that we have this nicely separated in different commits? See also another small suggestion below.

[Cito]

👍 Thank you.