feat(gateway): multi-platform delivery simulator

[JoaoMarcos44]

Summary

Adds hermes gateway simulate-delivery --platform <p> --text/--input [--json] — a preview command that renders and chunks content through the REAL adapter code for Telegram, Discord, Slack, and WhatsApp, with zero network calls and no credentials required.

%%{init: {'theme': 'dark', 'themeVariables': { 'primaryColor': '#8b0000', 'mainBkg': '#0a0204', 'primaryTextColor': '#ffccd5', 'primaryBorderColor': '#ff0038', 'lineColor': '#ff0038'}}}%%
graph TD
    A[Raw Input Text] -->|Real format_message| B[Adapter Renderer]
    B -->|Telegram| C[MarkdownV2 Escaping]
    B -->|Discord| D[GFM Table to Bullets]
    B -->|Slack| E[mrkdwn]
    B -->|WhatsApp| F[Native Formatting]
    C --> G[truncate_message Chunker]
    D --> G
    E --> G
    F --> G
    G -->|No Network, No Credentials| H[Formatted Output + Chunk Report]

Loading

Infographic :

Closes #80174

Why this doesn't duplicate existing work

• feat(conformance): vector generator — native renderers as executable spec #71666 (merged) is the CI conformance-vector generator — a fixed corpus rendered through the same oracles for regression testing. This is the ad-hoc/interactive counterpart for arbitrary developer-supplied content, reusing the identical oracle functions rather than reimplementing them.
• feat(gateway): declare per-adapter media-send capabilities #61694 (open) is media-capability metadata; this is text rendering/chunking — different surface.
• test(gateway): exercise bundled platform contracts #74953 (open) is a static conformance test suite; this is an operational preview tool, not a test.

Design notes

• Oracles imported, never reimplemented. format_message() is called unbound (AdapterClass.format_message(None, text)) for Telegram/Discord/Slack, and via a bare object.__new__(WhatsAppBehaviorMixin) instance for WhatsApp — the exact construction scripts/generate_conformance_vectors.py uses, asserted self-free by tests/conformance/test_vector_generator.py.
• Chunking is the real BasePlatformAdapter.truncate_message() static method (code-fence-boundary preservation included), called with each adapter's real MAX_MESSAGE_LENGTH and len_fn (Telegram measures UTF-16 code units via utf16_len, matching gateway/delivery.py's real call site).
• Scoped down from the original proposal. The original report also asked for draft-vs-final streaming diffs and logical wire payloads. Those are genuinely adapter-specific (Telegram edit-streaming, Discord multi-message threading) and reimplementing them "byte-for-byte" within a correctness-first MVP risks shipping a tool that lies to the user — worse than not having it. This PR ships the rendering+chunking slice only, which is also the exact bug class bug(telegram): rich_messages + draft streaming renders crooked MDV2 preview then a second wiki/rich final #78524 reports. Draft/streaming simulation is a natural follow-up once this lands.
• New CLI wiring: simulate-delivery subparser added to the existing gateway parser (hermes_cli/subcommands/gateway.py), dispatched via one guard clause added at the top of hermes_cli/gateway.py::_gateway_command_inner — no existing gateway subcommand branch was touched.

Test plan

• pytest tests/hermes_cli/test_gateway_delivery_simulator.py -q — 11 passing: real MarkdownV2 escaping, Discord pass-through, UTF-16 length check, chunk-boundary correctness (len(chunk) <= max_length for every chunk), unsupported-platform error, file-input path, JSON output
• pytest tests/hermes_cli/test_gateway.py -q — no regressions (10 passed, 5 pre-existing skips)
• ruff check on new/changed files — clean
• Manual run: hermes gateway simulate-delivery --platform telegram --text "Price 3.50 (was 4.00) save ~12%!" → correctly escaped MarkdownV2 output; --platform whatsapp --json; unsupported platform → clean error message