feat: prune tensor expr inspect results

[polvalente]

Adds pruning of in-between expressions so that we
it's easier to analyze the overall behavior of a Nx.Defn.Expr

[polvalente]

Currently this prints:

Nx.default_backend(Nx.Defn.Expr)
param_tensor = &Nx.Defn.Expr.parameter(Nx.Defn.Expr.tensor(&1), &2)
Enum.reduce(1..50, param_tensor.(0, 0), &Nx.add(&2, param_tensor.(&1, &1))) |> IO.inspect(limit: 10); nil
#Nx.Tensor<
  s64
  
  Nx.Defn.Expr
  parameter a:0     s64
  parameter b:1     s64
  parameter d:2     s64
  parameter f:3     s64
  parameter h:4     s64
  parameter j:5     s64
  parameter l:6     s64
  parameter n:7     s64
  parameter p:8     s64
  ...               
  parameter dv:50   s64
  ...               
  de = add dc, dd   s64
  dg = add de, df   s64
  di = add dg, dh   s64
  dk = add di, dj   s64
  dm = add dk, dl   s64
  do = add dm, dn   s64
  dq = add do, dp   s64
  ds = add dq, dr   s64
  du = add ds, dt   s64
  dw = add du, dv   s64
>

[jonatanklosko]

Suggested change

	(len = length(parameters)) > opts.limit ->
	[h | t] = parameters
	[h, {"...", ""} | Enum.drop(t, len - opts.limit)]
	length(parameters) > opts.limit ->
	[h | t] = parameters
	[h, {"...", ""} | Enum.take(t, -(opts.limit - 1))]

[josevalim]

@polvalente sorry, I pushed something but I messed up.

[josevalim]

@polvalente it shuold be good now. I used your idea of appending/prepending to the parameters and using the limit as a flag of when to do that once, so we don't need to do a later check. We just need tests and we are good. :)

[polvalente]

@polvalente it shuold be good now. I used your idea of appending/prepending to the parameters and using the limit as a flag of when to do that once, so we don't need to do a later check. We just need tests and we are good. :)

Awesome! I'll work on testing this :)

[polvalente]

@josevalim tests are kinda verbose, but I wanted to cover most edge cases