Engine: CEL evaluator core — value model over DynamicValue, operators, short-circuit, macros/comprehensions

[rrrodzilla]

Workstream: Engine · Part of #89

Context

Tree-walking evaluator over the parser AST (#107), producing CEL values over our DynamicValue domain.

Scope

• CEL value model bridged to DynamicValue, including CEL-internal types not yet in the DSL (uint, bytes, duration) — coordinate with type issues Type: duration (records retention / TTL / SLA) + timestamp arithmetic #96/Type: bytes (inline binary — hashes, signatures, key material) #97/Type: uint — distinct type vs unsigned constraint (decision) #98/Type/semantics: CEL timestamp operations (tz-aware accessors, overflow rules) #101.
• Operator semantics: arithmetic, comparison, logical with short-circuit (&&, ||, ternary) and CEL error/absent propagation rules.
• Macros / comprehensions: has, all, exists, exists_one, map, filter.
• Deterministic, no I/O, guaranteed-terminating (bounded comprehension semantics) — these are the properties that justify the whole rules-engine thesis (Epic: Built-in declarative rules engine (reduce reliance on gRPC hooks) #89).

Acceptance

• Passes the basic, logic, comparisons, lists, macros/macros2 subset sections of Spike: measure pure-Rust CEL conformance against the official cel-spec suite #90.

---

Dependencies: Blocked by #107 (parser). Blocks the stdlib and the capabilities (#92/#93/#94). Hard.

[rrrodzilla]

Inherited from #107 (parser) — error-absorption semantics for all/exists.

The parser lowers iteration macros to Comprehension nodes at parse time (commit a675303). For all/exists, the loop_condition is currently the plain accumulator (@result / !@result).

Canonical CEL wraps the loop condition in an internal @not_strictly_false(...) so that an error in the predicate on one element does not abort the whole macro when another element already determines the result — e.g. [1, 'x', 3].exists(e, e == 2) should still short-circuit/absorb per spec rather than propagate the type error unconditionally. This is an evaluator-semantics concern, not a parse concern, so it was deliberately deferred here.

When building #108 test-first against the macros/macros2 conformance sections:

• Decide whether to implement absorption in the comprehension evaluator directly, or to re-introduce the @not_strictly_false wrapper in the lowering (crates/schema-forge-cel/src/parser.rs::macros).
• Verify the exact internal function spelling against the corpus expectations before baking it into the AST.

The macro lowering functions are pure and isolated in the private macros submodule, so adjusting the loop_condition shape is a localized change.

[rrrodzilla]

Landed in 31c531e — feat(cel): add tree-walking evaluator core (#108). The evaluator walks the AST against a Scope over the owned CelValue model (operators, short-circuit, ternary, and the lowered comprehension macros). Closing; code merged on dev.