Fix/minimal space owned eval 20260827

[zariuq]

Two weird behaviors detected:

1. Types adedd after a function may not apply to it
2. Space ownership of duplicate terms led to odd behavior (duplicates, seeming removal from space, etc).

See this program:

; Store the same equation under two independent owners.
!(bind! &foo (new-space))
!(add-atom &self (= (f) 1))
!(add-atom &foo (= (f) 1))

; Only &self contributes to default-context execution.
!(collapse (f))

; Remove only the copy owned by &foo.
!(remove-atom &foo (= (f) 1))

; Both &self's data and executable behavior remain intact.
!(match &self (= (f) $result) $result)
!(collapse (f))

PeTTa's old behavior:

true
true
(1 1)
true
1
((f))

PeTTa's new behavior:

true
true
(1)
true
1
(1)

Codex and Claaude's solution is tog ive each space ownership of its compiled program, adding eval/2 for explicit context denotation (e.g., (eval $expr $space)), etc.

As to the actual PR, the edits are pretty extensive, so I might suggest using them as a reference for your own fix (if you agree with these semantics).

I think the coherent options are:
A) Something like this.
B) Something like MORK with one universal space.

The current version seems 'incoherent' (in weird cases).