A new method for the greedy optimizer

[ArrogantGao]

The following changes have been made in this PR:

1. Added a new optimizer "Hyper-Greedy". In each step, instead of directly selecting the contraction pair with the lowest cost, here we sampling one of them according to the cost $\exp(-L / t)$, where $L$ is the cost and $t$ is the temperature, thus it may given better results that greedy.
2. Deps of BetterExp is removed, since it has been merge to Base since [email protected], and currently exp2 provided by Base seems better:

julia> @btime exp2(1.0);
  0.791 ns (0 allocations: 0 bytes)

julia> @btime BetterExp.exp2(1.0);
  1.625 ns (0 allocations: 0 bytes)

(betterexp) pkg> st
Status `~/temp/betterexp/Project.toml`
  [6e4b80f9] BenchmarkTools v1.5.0
  [7cffe744] BetterExp v0.1.0

3. Added multi-threads support for tree_greedy.

[GiggleLiu]

Add parameters alpha, move temperature to greedy.

[GiggleLiu]

temperature

[ArrogantGao]

Added a new struct GreedyStrategy for method of GreedyMethod.
GreedyStrategy include parameters $\alpha$ and temperature, thus MinSpaceOut and MinSpaceDiff now are special case of GreedyStrategy.

[GiggleLiu]

GreedyStrategy seems to be not nessesary. We can move the fields to GreedyMethod.

[GiggleLiu]

vi and vj not defined here. How about:

We use number 0, 1, 2 to denote the output tensor, the first input tensor and the second input tensor,
and use e.g. `l01` to denote the set of labels that appear in both the output tensor and the input tensor.

[GiggleLiu]

Suggested change

	# eincode = optimize_code(parse_eincode(incidence_list, tree, vertices=1:length(parts)), log2_sizes, sub_optimizer)
	# if eincode isa SlicedEinsum
	# @assert eincode.slicing == 0
	# eincode = eincode.eins
	# end
	# tree = parse_tree(eincode, collect(1:length(parts)))

[GiggleLiu]

Suggested change

# res = optimize_greedy(inset, iy1, size_dict; method=greedy_config.method, nrepeat=greedy_config.nrepeat)

[GiggleLiu]

Suggested change

	# eincode3 = ein"(ab,acd),bcef,e,df->"
	# Random.seed!(2)
	# optcode3 = optimize_greedy(eincode3, size_dict)
	# tc, sc = timespace_complexity(optcode3, edge_sizes)
	# @test 16 <= tc <= log2(exp2(10)+exp2(16)+exp2(15)+exp2(9)+1e-8)

[ArrogantGao]

These commented parts have been removed, and the doc strings are revised.