Fix quadratic runtimes in Mathics

[poke1024]

This is continuation of #575 and #615. It's my cleaned up proposal for fixing quadratic runtimes in Mathics.

The following plots show the results of the accompanying two commands, run once using master (yellowish) and once using this PR (blue). Don't ask me where the spurious regular dots come from (garbage collection?).

First[Timing[Fold[#1+#2&, Range[#]]]]& /@ Range[750]

MyTotal[x0_] := Module[{x = x0, y = 0}, While[Length[x] > 0, y = y + First[x]; x = Rest[x]]; y];
First[Timing[MyTotal[Range[#]]]]& /@ Range[500]

The changes are twofold:

• First, a new class called a Structure was introduced to enable list operations like Most or Take not to trigger an evaluation of every leaf again (this used to be Faster Partition[] #615).
• Second, the cache check is now based on a set of symbols rather than a full tree traversal (this used to be Roughly constant time expression timestamp checks #575).

The quadratic runtimes are only fixed when both measures are applied, which is why I merged everything into this PR now.

All the information regarding cached Expression information (including what is Expression._sequences in master) is now bundled in ExpressionCache, which makes the code much cleaner I think.

I experimented with quite a variety of data structures, including dicts instead of sets for the symbol caching, but a set and a sequences list seems to be the best thing to do.

[GarkGarcia]

@mmatera This seems related to #844

[mmatera]

I am trying to do a rebase of this PR, but there are something that I do not completely understand: in Expression, in the PR, leaves are represented by tuples. This originates some errors in operations that try to modify the leaves of an expression. I do not understand the motivation for using tuples instead of lists. Is about safety in the code? or about performance? @poke1024 could you give me a hand with this?

(please see the merge at https://github.com/mmatera/Mathics/tree/qtime)

[rocky]

#936 should be the same thing rebased on current sources.