Prototype exact probability distributions for the v2 language

## Parent

- #3

## What to build

Build a decision-focused prototype that computes exact distributions for the provisional RFC-0006 subset and measures state growth for modifiers, arithmetic, and nested expressions.

## RFC-0006 prototype contract

- Consume `docs/rfcs/0006-capability-matrix.json`; do not expand support implicitly.
- Produce self-contained AnalysisValue outcomes with reduced arbitrary-precision rational probabilities.
- Reproduce every case in `docs/rfcs/0006-distributions.json` by exhaustive enumeration.
- Never invoke the random evaluator and never fall back to Monte Carlo.
- Reject possible runtime errors atomically without a partial distribution.
- Record states visited, peak states, support size, operations, estimated bytes, rational bit width, and elapsed time.
- Exercise the proposed defaults and hard maxima in `docs/rfcs/0006-analysis-contract.json`.
- Publish evidence for every construct currently marked `bounded`.

## Acceptance criteria

- [ ] The prototype produces exact distributions for the proposed baseline subset.
- [ ] All normative tables are reproduced independently.
- [ ] Complexity and memory growth are measured for each candidate language feature.
- [ ] Every bounded construct receives keep/narrow/reject evidence.
- [ ] Unsupported constructs and viable safety ceilings are documented for #33.
- [ ] Prototype tests are deterministic and run in the RFC-0005 conformance lifecycle.

## Blocked by

- #3 — Draft contract delivered; final review follows this prototype.
- #16
- #15
- #24

Start by reading docs/rfcs/0006-capability-matrix.json, docs/rfcs/0006-distributions.json, and docs/rfcs/0006-analysis-contract.json, then review the RFC-0005 conformance lifecycle. The prototype is done when normative distributions are reproduced deterministically, bounded constructs have complexity and memory evidence, unsupported constructs and safety ceilings are documented for #33, and tests run in that lifecycle.

Written by the indexing model from the issue text.