Add lazy option to QuantumOpticsRepr

[GrovyleX]

What this does

Adds an opt-in lazy::Bool=false field to QuantumOpticsRepr. On its own this does nothing
visible — it's the switch the companion QuantumSymbolics PR (QuantumSavory/QuantumSymbolics.jl#203) reads to decide whether
express builds eager matrices or QuantumOpticsBase's lazy operators (LazySum, LazyProduct,
LazyTensor). Together they implement qojulia/QuantumOptics.jl#521. The default is unchanged:
QuantumOpticsRepr() is still eager.

Keeping the existing constructors working

QuantumOpticsRepr is a Base.@kwdef struct, so naively adding a second field breaks the
positional call QuantumOpticsRepr(4) — it would start expecting two positional arguments. One
outer constructor restores it:

QuantumOpticsRepr(cutoff::Int) = QuantumOpticsRepr(; cutoff)

So nothing existing breaks and the option is purely additive:

• QuantumOpticsRepr() → cutoff 2, eager
• QuantumOpticsRepr(4) → cutoff 4, eager
• QuantumOpticsRepr(cutoff=4)
• QuantumOpticsRepr(lazy=true)
• QuantumOpticsRepr(cutoff=4, lazy=true)

Why a field, and not a flag passed some other way

express caches its conversions in a Dict keyed on the representation value, and
QuantumOpticsRepr relies on the default structural ==/hash. Making lazy a real field means
an eager and a lazy representation of the same object compare unequal, so they get separate cache
slots and the cache can't hand back the wrong type. test/test_express.jl checks that invariant
plus all the constructor forms above.

Pkg.test() is green locally on Julia 1.12.6, including Aqua. CHANGELOG bumped to v0.4.3.

Part of qojulia/QuantumOptics.jl#521

AI disclosure

I used AI to help analyze the codebase understand different and help in implementation, but I worked out the @kwdef/constructor and caching details myself, verified them against how express actually keys its cache, and wrote
and ran the tests. I went through the whole diff and understand every line - happy to walk through
any of it.