unify prod(::Tuple{Vararg{Int}}) with a single method

[nsajko]

One less method.

[Zentrik]

This has caused quite a few performance regressions in the array indexing benchmarks in BaseBenchmarks. See for example the following regressions taken from https://github.com/JuliaCI/NanosoldierReports/blob/master/benchmark/by_date/2025-03/21/report.md and https://tealquaternion.camdvr.org/compare.html?start=9294132415cfdf2198b64d03f0fd58f8a4d3a258&end=b8dc57e59db11658cb57424e4e12773f572a6517&stat=min-wall-time&name=array.index

Benchmark	% Change
array.index.ind2sub	927.68%
array.index.(mapr_access, SubArray{Float32, 2, BaseBenchmarks.ArrayBenchmarks.ArrayLS{Float32, 3}, Tuple{Int64, Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}}, false})	514.02%
array.index.(mapr_access, SubArray{Float32, 2, Array{Float32, 3}, Tuple{Int64, Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}}, true})	481.90%
array.index.(sumlinear_view, Base.ReinterpretArray{Float32, 2, Int32, Matrix{Int32}, false})	449.60%
array.index.(sumeach_view, Base.ReinterpretArray{Float32, 2, Int32, Matrix{Int32}, false})	414.18%
array.index.(mapr_access, SubArray{Int32, 2, Array{Int32, 3}, Tuple{Int64, Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}}, true})	317.28%
array.index.(mapr_access, SubArray{Int32, 2, BaseBenchmarks.ArrayBenchmarks.ArrayLS{Int32, 3}, Tuple{Int64, Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}}, false})	295.44%
array.index.(sumvector_view, Base.ReinterpretArray{BaseBenchmarks.ArrayBenchmarks.PairVals{Float32}, 2, Float64, Matrix{Float64}, false})	193.03%
array.index.(sumlinear_view, Base.ReinterpretArray{BaseBenchmarks.ArrayBenchmarks.PairVals{Int32}, 2, Int64, Matrix{Int64}, false})	172.68%
array.index.(sumeach_view, Base.ReinterpretArray{BaseBenchmarks.ArrayBenchmarks.PairVals{Int32}, 2, Int64, Matrix{Int64}, false})	168.50%
array.index.(sumvector_view, Base.ReinterpretArray{BaseBenchmarks.ArrayBenchmarks.PairVals{Int32}, 2, Int64, Matrix{Int64}, false})	156.48%
array.index.(sumvector_view, Base.ReinterpretArray{Float32, 2, Int32, Matrix{Int32}, false})	147.77%

[nsajko]

Observations regarding the regression:

• code_llvm and code_native looks fine and mostly unchanged for prod
• the inlining cost for prod is now different because of the loop, as shown by Base.print_statement_costs
• profiling run(BaseBenchmarks.SUITE[["array", "index", "ind2sub"]]) doesn't show either prod or * in the flame graph (perhaps they're inlined after all?)
• adding @inline to prod doesn't help