v0.2.2 - Optimisations and stabilisation

Bugs fixed since v0.2.1:

• Fixed the while loop
  The test for the while loop was inverted
  The location of the end and repeat keywords was swapped

• Fixed bug in detection of multiplication overflow
  The computation of the factorial of 23 and up is now correct.

• Fixed bug mistakenly converting some fractions to integers
  For example, 2/17 INV would return 17 instead of 17/2

• Add spacing to large numbers
  A number that did not fit on 64-bit would not respect spacing settings

• Repaired the test suite
  Tests were setting-dependent, and would fail with default settings because
  they did not take number spacing into account.

Features implemented since v0.2.1:

• Implemented local variables, e.g. << -> A B C << A B - C * >> >>
  Local variables are a very fast way to reference data. Internally, they are
  represented by an offset to a known location, which makes them practically as
  fast as stack accesses, and much faster than global variable lookups.

• Implement the same function
  The same function detects if two objects are identical.
  Like the HP48, there are multiple "equality" variants:

  • same detects if two objects are binary-identical
  • == does the same, but dereference names
  • = is a mathematical equality (does not exist in HP48 RPL)

• Add feature to import ASCII versions of «, » and →
  In .48S files, you can use <<, >> and -> instead.

• Implement →List (stack to list)
  This was the last missing command to be able to execute the
  HP calculator's N-queens benchmark

• Better truncation of very large numbers when shown on stack
  A very large number, like the factorial of 64, would show up truncated
  at 80 digits, so the first digits shown were wrong.

• Various optimizations to make things a bit snappier, notably on battery power

  • Refactored the object model to use a simpler lookup mechanism

  • Implemented a C++ interface for arithmetic and functions, which will be
    useful when implementing complex number and matrix operations, and already
    simplified the code for operations such as abs or cubed.

  • Sime fine-tuning of LEB128 handling for types

  • Execution time for NQueens measured, reduced from about 980ms down
    to about 530ms.

Full Changelog: v0.2.1...v0.2.2