Silent Slide drop on missing separator

Start in src/presentations/spec_parser.py, especially _split_slides_fence_aware and _parse_slide, then run the checked-in mse.py reproduction with uv run. Done means consecutive slide headers are split into separate slides, the three-slide example renders all three, and a warning identifies the header missing its separator.

Summary

A spec that defines N slides renders only N−1, with no warning, when the author omits a --- separator between two ## [type] headers. The second slide's heading and bullets are silently swallowed into the first slide's speaker notes.

Impact

High — a one-character authoring typo causes silent, unreported data loss (a whole slide disappears). Affects all spec parsing.

Environment

• ms-presentations version: 0.1.6
• Python version: 3.10+
• OS / version: reproduces on all (pure parser logic; verified on Windows 11)
• Install method: pip / from source
• Relevant deps (if Azure path): n/a — parser only, no Azure
• Azure / model config: n/a

Steps to Reproduce

1. Write a spec with two consecutive ## [type] slides but no --- between them.
2. Parse/render the spec.
3. Observe the deck has one fewer slide than authored, with no warning.

Minimal Reproducible Example (MRE)

The script below is also checked in as mse.py in this folder.

# /// script
# requires-python = ">=3.10"
# dependencies = ["ms-presentations==0.1.6"]
# ///
"""Self-contained MRE — Silent slide drop on a missing '---' separator.

Run with:  uv run mse.py   (no venv, no Azure, no network)

When the '---' separator between two '## [type]' headers is missing, the second
slide is folded into the first slide's notes and dropped. The spec below defines
three '## [content]' slides; the rendered deck contains only two. The dropped
slide is the observable error — the deck itself is the only output.
"""
import os
import tempfile

from presentations.cli import main

SPEC = """\
---
title: Repro - Silent slide drop
output: Silent_Slide_Drop.pptx
---

## [content] Slide One (renders)

- Properly separated by a --- below.

**Notes**: First slide.

---

## [content] Slide Two (absorbs slide three)

- This slide is followed by a '## [content]' header below,
- but there is NO '---' separator between them.

**Notes**: Everything below, including Slide Three, gets swallowed here.

## [content] Slide Three (silently dropped)

- These bullets never become their own slide.

**Notes**: Expected 3 slides, actual 2.
"""

with tempfile.TemporaryDirectory() as d:
    spec_path = os.path.join(d, "minimal.spec.md")
    with open(spec_path, "w", encoding="utf-8") as f:
        f.write(SPEC)

    main([spec_path, "-o", os.getcwd()])  # renders the deck into the PWD

uv run mse.py

Expected Behavior

The parser produces 3 slides — and ideally emits a warning naming the header that lacked a separator.

Actual Behavior

parse_spec returns 2 slides; "Slide Three" is absorbed into "Slide Two" notes. No warning is printed and rendering succeeds, hiding the loss.

Additional Context

Root cause: _split_slides_fence_aware (src/presentations/spec_parser.py:40-57) splits only on --- lines. With no separator both headers land in one chunk; _parse_slide (spec_parser.py:81) matches the first header and treats everything after the first **Notes**: as that slide's notes.

Suggested fix: when a second ## [type] header appears in a chunk that already has one (and not inside a code fence), start a new chunk and print a warning naming the offending header. Fully isolated — different file, no shared state with any other fix.