Promote per-adapter card helpers into `shared/adapter_utils.py` (emoji converter, button styles, GFM table, fallback text)

[patrick-chinchill]

Code-organization gap — drift risk across 8 adapters.

Current state

src/chat_sdk/shared/adapter_utils.py exposes only extract_card() and extract_files(). Every other cross-adapter helper lives duplicated inside each adapter's cards.py / format_converter.py:

• Emoji placeholder resolution ({{emoji:thumbs_up}} → platform-native code). Each adapter imports DEFAULT_EMOJI_MAP from shared/ and walks children independently. Slack, Discord, Telegram, Teams, Google Chat, Linear, GitHub all do this; WhatsApp doesn't (because it doesn't translate emoji placeholders at all — consistent gap).
• Button style mapping (style: "primary" | "secondary" | "danger" → Block Kit "primary" / AdaptiveCard "positive" / Discord 1..4 / Linear mention syntax). Scattered across each adapter.
• Card fallback text (render a Card to plain markdown for platforms with limited card support, or for notification previews). Reinvented in Telegram and WhatsApp; partially in Discord.
• GFM table rendering (render Table to an ASCII-aligned markdown table for adapters without native table blocks). Duplicated in Telegram, WhatsApp, Linear; Discord has its own; Slack has a separate Block-Kit-first path.
• Table cell escaping (|, newlines, backticks inside cells). Each adapter that renders tables has its own escape routine; subtle bugs have bitten us in Slack tables specifically.

Why it matters

• Drift: eight adapters solving the same problem independently means eight different bug surfaces for the same logical operation. We've already shipped one fix to Slack empty-header cells (0.4.26) that didn't propagate to other adapters because there was no shared routine to fix.
• Maintenance: new platform emoji added to DEFAULT_EMOJI_MAP requires audit of every adapter to confirm it's picked up. A shared converter removes the audit.
• Consistency: "render this Card as fallback text" should produce the same output regardless of which adapter is asking. Today it doesn't.

Proposed surface

In src/chat_sdk/shared/adapter_utils.py:

def create_emoji_converter(
    platform_map: Mapping[str, str] | None = None,
) -> Callable[[str], str]:
    """Return a function that replaces {{emoji:name}} placeholders with platform-native code."""

def map_button_style(
    style: ButtonStyle,
    target: Literal["slack", "teams", "discord", "telegram", "whatsapp", "gchat", "linear", "github"],
) -> str | int:
    """Translate SDK button style to the target platform's native representation."""

def card_to_fallback_text(card: Card, *, emoji_converter: Callable[[str], str] | None = None) -> str:
    """Render a Card to plain markdown text, honoring emoji placeholders."""

def render_gfm_table(table: Table, *, aligned: bool = True) -> str:
    """Render a Table to a GFM-flavored markdown table, optionally padding for visual alignment."""

def escape_table_cell(text: str) -> str:
    """Escape `|`, newlines, and backticks for safe inclusion inside a markdown table cell."""

Each helper replaces the per-adapter copy. Per-adapter code stays responsible for (a) choosing which helpers to call, (b) mapping to platform-specific payload shapes (Block Kit, AdaptiveCard, Discord components, etc.) — that's legitimately platform-specific and shouldn't move.

Migration

One PR per helper, in this order:

1. escape_table_cell (most localized; low risk)
2. render_gfm_table (builds on security: Fix all critical and high findings from security audit #1)
3. card_to_fallback_text (uses security: Fix all critical and high findings from security audit #1 + fix: correct GitHub Actions commit SHAs #2)
4. map_button_style (independent)
5. create_emoji_converter (largest blast radius; save for last when the pattern is settled)

Each PR: promote the helper, update all adapters to use it, delete the duplicated copies, add regression tests in shared/ for the helper itself + keep existing adapter tests passing.

Acceptance

• All five helpers live in shared/adapter_utils.py with direct unit tests
• Each of the 8 adapters calls the shared helper instead of a local copy
• Each migration PR shows net LOC deletion
• No behavior change (captured by existing adapter tests; add a parity test that runs the same Card through every adapter's fallback path and snapshots the output)