PROMPT 04 Ready Failures And Retry

# Implementation Prompt 04: Ready State - Failures and Retry Logic

**Goal:** Implement :ready state failure handling, transport busy retry, oversized frames, and transport down handling (reset notifications were removed from MVP).

> **Update (2025-11-07):** Use `:erlang.send_after/3` for retry timers (message form `{:retry_send, id, frame}`) and ignore any legacy instructions about `{:info, {:retry_send, ...}}`. Reset notifications are intentionally out-of-scope for MVP; the remaining references in this prompt are historical context only.

**Test Strategy:** TDD with rgr. All tests green, no warnings.

---

## Context: What You're Building

You're implementing the failure paths and retry logic for :ready state:
1. **Transport busy retry**: When send returns `:busy`, schedule retry with exponential backoff + jitter
2. **Transport down**: Fail all in-flight and in-retry requests, transition to :backoff
3. **Oversized frames**: Close transport, transition to :backoff
4. **Reset notifications**: (Removed for MVP; keep this slot empty until a negotiated capability exists)
5. **Stop during retry**: Clear retry timers, prevent double replies

This is where **concurrent retry correctness** is critical (per-ID retry tracking).

---

## Required Reading: State Transitions

From STATE_TRANSITIONS.md, these are the exact transitions to implement:

### :ready State - Retry Path

| From | Event | Guard | Actions | To |
|------|-------|-------|---------|-----|
| :ready | `{:call, from}, {:request, method, params, opts}` | send returns `:busy` | Generate ID; track in `retries` map with attempt count; schedule retry with jitter; reply `{:ok, id}` | :ready |
| :ready | `{:info, {:retry_send, id}}` | retry exists, attempts < max | Increment attempts; retry send; if `:ok` → promote to requests with timeout; if `:busy` → reschedule retry; if error → fail request | :ready |
| :ready | `{:info, {:retry_send, id}}` | retry exists, attempts ≥ max | Reply backpressure error; tombstone; clear retry | :ready |
| :ready | `{:info, {:retry_send, id}}` | retry cleared (stop) | Ignore | :ready |

### :ready State - Failure Paths

| From | Event | Guard | Actions | To |
|------|-------|-------|---------|-----|
| :ready | `{:info, {:transport, :down, reason}}` | - | Tombstone all requests; **fail/clear retries**; do not re-arm set_active; schedule backoff | :backoff |
| :ready | `{:info, {:transport, :frame, binary}}` | byte_size > max | Log error; close transport (no set_active); **fail/clear retries**; schedule backoff | :backoff |
| :ready | `{:call, from}, :stop` | - | Tombstone all requests; **fail/clear retries**; reply `:ok`; close transport; exit(normal) | - |

### Retry Tracking Structure (from ADR-0007)