fix(deps): pin markitdown >=0.1.5 with narrow extras (closes #64)

[KylinMountain]

Summary

• Replace markitdown[all] with markitdown[docx,pptx,xlsx,xls]>=0.1.5 — drops ~67 MB of unused deps (azure-*, pdfminer, pdfplumber, speechrecognition, youtube-transcript-api, pydub, xlrd-then-readded, olefile) and forces users off pre-0.1.0 markitdown.
• Add .xls to SUPPORTED_EXTENSIONS / _SHORT_DOC_TYPES so the [xls] extra is actually reachable.

Why this closes #64

The reporter's traceback (_markitdown.py:1715) is from the pre-0.1.0 monolithic markitdown, where a ValueError from python-pptx's CT_StrVal_NumVal_Composite.value (bare float(self.v.text) on #N/A / #DIV/0! / blank cells) propagates straight out and aborts the whole openkb add run.

From markitdown 0.1.2 onward, _convert_chart_to_markdown is wrapped in except Exception and the offending chart degrades to [unsupported chart]. Pinning >=0.1.5 guarantees users land on a version with that handler.

Caveat

This is an "unblock the pipeline" fix, not a chart-data-preservation fix — the chart's other (numeric) cells are still lost on bad cells. The root cause sits in python-pptx and there is no upstream fix or open PR there; we considered a monkey-patch but dropped it as too invasive. A higher-fidelity pptx path (e.g. LibreOffice → PDF → existing PDF pipeline) behind a config flag is a reasonable follow-up if users hit this in practice.

Test plan

• uv run --extra dev pytest → 520 passed
• uv sync after the dep change cleanly removes 17 packages, then re-adds xlrd for [xls]
• Manual: from openkb.converter import MarkItDown still imports

[KylinMountain]

Code review

No issues found. Checked for bugs and CLAUDE.md compliance.

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}