Drop the redundant literal_pow forwards

[jishnub]

Depends on JuliaArrays/FillArrays.jl#385 (FillArrays 1.18) — not on #444. Nothing here calls the new simplify_broadcasted entry point, so the compat bound is raised only to FillArrays = "1.18", and this can land a step ahead of the rest of the chain. CI will not resolve until #385 is merged and released.

Why

infarrays.jl carried four methods forwarding x .^ k on infinite fills to DefaultArrayStyle, one for the all-infinite axis shape and three for the mixed-2D and square-infinite ones, purely to retain the FillArrays simplification.

They are now redundant twice over:

• FillArrays attaches its literal_pow rules to the operation rather than to a style (as of #385), so x .^ k is simplified before any style is consulted;
• LazyArrays' generic forward already covers LazyArrayStyle{N} against AbstractFill{T,N} with no axis constraint.

Verification

Rather than trusting the test suite alone, I checked the specific shapes. Against FillArrays 1.18 with LazyArrays master, and with these four methods deleted, the literal_pow testset still passes across 1D, 2D all-infinite, both mixed-2D axis orders and 3D:

Ones(∞) .^ 2 ≡ Ones(∞)          Fill(2,∞) .^ 2 ≡ Fill(4,∞)
Ones(ax...) .^ 2 ≡ Ones(ax...)  for ax in ((∞,∞), (∞,5), (5,∞))
Ones(∞,∞,∞) .^ 2 ≡ Ones(∞,∞,∞)

A behavioural probe over those shapes plus Ones(∞) .* (1:∞), Fill(2,∞) .* (1:∞) and muladd.(Ones(∞), Ones(∞), Ones(∞)) is unchanged.

Full test suite passes.

🤖 Generated with Claude Code