Update readme with usage demo and some other context

Ultimately, many cursors may want to simultaneously hold a borrow to the pager.

• For reads, external behaviour of pager is immutable, but internal state must change to facilitate reading pages from the disk cache, or lazily allocating new empty pages in memory
• For writes, a unique/mutable borrow is of course needed, but only one page need be borrowed at a time

To achieve this, the PageFile and each individual Page will be placed in a RefCell.

Pages do not track number of populated rows, that comes from the single table object right now. But when we read back from disk, the number of rows is calculated as num_pages * ROWS_PER_PAGE, so the partial page is read back as a full page with many empty rows.

We could fix this by subtracting from row count based on size of zeroed memory at the end of the last page. However, it's not really worth the effort since we will soon be switching page format from contiguous arrays to btree nodes #4. Will mark the failing test as skip for now, and ensure it's passing with the btree implementation.

• add logging at various levels with the log crate
• convert existing log lines using eprintln! to use logging macros
• fix rexpect tests currently broken by logging output by setting log level appropriately

As the final task for the MVP milestone, I'd like to review the code and restructure/refactor as appropriate to clean things up ready for future changes.

• optional file path argument to repl binary
• pager abstraction between database operations and the underlying memory
• support for reading/writing pages to disk
• fall back to memory only if no path is provided
• additional rexpect tests to assert new behaviour works correctly in REPL

As per part 6 of the guide we're following, we want to put a cursor abstraction between the pager and the SQL queries.

Blocked by page format change #4