Spreadsheet support: .xlsx and .csv

[Phantom-VK]

Problem

Users working with data often want to count tokens in spreadsheets. Currently refused.

Design question

How a sheet becomes text changes the token count: cell-by-cell, row-joined, or Markdown tables. Whatever is picked should be documented, since the number depends on it.

Implementation

• parsing.py needs a _read_xlsx handler using openpyxl
• CSV can use the existing _read_text (already works)
• Decision: how to flatten the sheet, then update docs/DATA_SOURCES.md to explain it

Scope

Two separate files, two separate PRs.