feat: natural language

[stefan-gorules]

Natural language projection for expressions

This PR allows engine to describe expressions in a natural language. It adds a read-side API that turns any ZEN expression into a stream of structured tokens that a client can render as a plain sentence, in any language.

What it delivers

Some examples of expressions and the sentences a client renders from the projected tokens:

Expression	Shown to the user as
customer.age >= 18	customer.age is at least 18
customer.country in ['US', 'CA']	customer.country is one of US, CA
age in [18..65]	age is between 18 and 65
> 100 (decision table cell)	greater than 100
all(order.items as item, item.price > 100)	every order.items has price greater than 100

The engine does not produce those words. It produces tokens, and the words come from the client's dictionary:

• Token projection. An expression comes back as typed tokens (field, operator, number) instead of text. Each token carries its position in the original source, so edits made on the sentence can be written back precisely.
• No English in the engine. Tokens carry stable symbols ("gte", "where", "sum"). Clients map symbols to phrases.
• Editing guidance built in. Where it knows more than the raw text shows, the engine says so: enum fields include their allowed values with ready-to-insert literals, operators include the alternatives valid for the operand types, date-like strings are marked for date pickers.

How it helps

• It is the foundation for rule editors that show and edit logic as sentences, with the engine guaranteeing the sentence always matches the real expression.
• Type-aware hints let editors offer only valid choices, which prevents malformed rules at the source.
• Purely additive: evaluation, tracing and all existing APIs are unchanged.