Remove eval from ArtifactInfo#to_hash

[tas50]

Summary

ArtifactInfo#to_hash called eval(attribute) once per attribute:

def to_hash
  Hash[ATTRIBUTES.map { |attribute| [attribute.to_sym, eval(attribute)] }]
end

So reading 13 instance variables compiled 13 fresh Ruby ASTs on every call — 33 µs and 67 allocations to do the work of 13 ivar reads. #initialize had a smaller version of the same problem, building "@#{attribute}" and attribute.to_sym per attribute per object.

This precomputes the attribute name and instance variable name as Symbols once at load time and looks them up directly.

Measurements

Ruby 4.0.6:

	before	after	speedup	allocations
#to_hash	33194 ns	1814 ns	18.3x	67 → 2
#clone_with	37218 ns	3251 ns	11.4x	84 → 6
.new	2945 ns	1211 ns	2.4x	14 → 1

The gain compounds through Backend::Base#windows_artifact_fixup!, which calls #clone_with for every Windows artifact plus every supported desktop version:

	before	after	speedup
clone a 20-artifact set	744 µs	64 µs	11.7x

Also

Removing the eval drops a dynamic code path from a method that only ever needs to read instance variables. It is not exploitable today since ATTRIBUTES is a frozen literal, but it is a sharp edge worth not having in a method that is otherwise trivial.

Verification

• 576 unit examples and the functional spec pass; chefstyle clean.
• Explicit round-trip assertions: to_hash equals the input hash, clone_with overrides only the given keys, the generated readers agree with to_hash, and absent keys still read back nil.
• All 176 golden SHA256 digests match main.