How to access OpenXML content by page number?

You cannot reference OOXML content via page numbering at the OOXML data level alone.

• Hard page breaks are not the problem; hard page breaks can be counted.
• Soft page breaks are the problem. These are calculated according to
  line break and pagination algorithms which are implementation
  dependent; it is not intrinsic to the OOXML data. There is nothing
  to count.

What about w:lastRenderedPageBreak, which is a record of the position of a soft page break at the time the document was last rendered? No, w:lastRenderedPageBreak does not help in general either because:

• By definition, w:lastRenderedPageBreak position is stale when content has
  been changed since last opened by a program that paginates its
  content.
• In MS Word’s implementation, w:lastRenderedPageBreak is known to be unreliable in various circumstances including
  1. when table spans two pages
  2. when next page starts with an empty paragraph
  3. for
     multi-column layouts with text boxes starting a new column
  4. for
     large images or long sequences of blank lines

If you’re willing to accept a dependence on Word Automation, with all of its inherent licensing and server operation limitations, then you have a chance of determining page boundaries, page numberings, page counts, etc.

Otherwise, the only real answer is to move beyond page-based referencing frameworks that are dependent upon proprietary, implementation-specific pagination algorithms.