Warn against possible code invalidations

Suppose I execute the following code in a fresh REPL (I haven't imported SpecialFunctions):

julia> f(x) = erf(x);
julia> f(1)
ERROR: UndefVarError: erf not defined
Stacktrace:
 [1] f(::Int64) at ./REPL[1]:1
 [2] top-level scope at REPL[2]:1
julia> using BenchmarkTools, SpecialFunctions
julia> x = 1; @btime f($x);
  27.645 ns (1 allocation: 16 bytes)

This allocation is not "real". See upstream issue: https://github.com/JuliaLang/julia/issues/35889. If I start a new fresh REPL, and this time load SpecialFunctions at the beginning:

julia> using BenchmarkTools, SpecialFunctions
julia> f(x) = erf(x);
julia> x = 1; @btime f($x);
  22.195 ns (0 allocations: 0 bytes)

the allocation is gone.

I think we should warn about this in the README or the docs.

Start with the README or documentation in BenchmarkTools.jl and review the invalidation example and linked Julia issue #35889. Add a warning explaining that defining code before loading SpecialFunctions can affect benchmark allocations, and confirm the documented guidance covers both REPL sequences shown.