fix(docs): handle NaN metadata cells in macro table generator

[bkowshik]

What

Fix a crash in scripts/generate_macro_table.py that breaks make html whenever a dataset has a missing (NaN) value in an optional metadata column (country, DOI, etc.).

Why

pandas stores missing cells as float NaN, which is truthy. The string formatters in _format_cell guard with if not value: — that guard lets NaN through, and the formatter then runs str/len operations on a float and crashes:

TypeError: object of type 'float' has no len()             # country_flag
AttributeError: 'float' object has no attribute 'replace'  # _doi_link

This surfaced after the metadata catalog (#868) added many optional columns that are empty for most datasets.

Fix

Normalize NaN to "" once at the top of _format_cell, before dispatching to any formatter. This centrally covers every string formatter (country, DOI, data URL, health) instead of patching each one individually.

Verification

make html-noplot builds green (exit 0); the macro table generates for all 158 datasets.