fix several issues + add mdscale-based initialization

[devernay]

• restore original variable names to be consistent with the reference MATLAB version (https://github.com/tompollard/sammon/commit/b69b203b89b7bd322694fdae97a9033779d6140f)
• Sammon mapping is designed to run only if there are no duplicate data points, so sammontest removed duplicate data (https://github.com/tompollard/sammon/commit/e03ed608d3093ab78ee687e352cc0654e2b4aa7e)
• check that input data contains no duplicates (which result in zero elements in the dissimilarity matrix) and better error reporting (https://github.com/tompollard/sammon/commit/5d674cca78ad8b77d0d9c26234c8229e4fcaa446 https://github.com/tompollard/sammon/pull/8/commits/1e923246d69a321015e7269b0aa8e447a2b6e8c7 https://github.com/tompollard/sammon/pull/8/commits/8791d5958b1b99bbe0e6b5d2c808b7e3c0133988)
• remove hacks that set infinity values to 0 in the dissimilarity matrix (https://github.com/tompollard/sammon/commit/ed05bc67fa87ad396f14d03bd2c602b7ed40575c)
• use scipy.spatial.distance.cdist instead of euclid (from previous PR) (https://github.com/tompollard/sammon/pull/8/commits/a677473b9f767c8297f1c7a57e49cff40904156f)
• fix bug when n != 2 (https://github.com/tompollard/sammon/pull/8/commits/a5852922e3e169cb14645de9e27d417e6cb6f755)
• add support for cmdscale-based initialization, which can be used when the input is a dissimilarity matrix (https://github.com/tompollard/sammon/commit/21bbdd6a08e99c5f8cccd4bc44e1d4468234f552)

[tompollard]

@devernay thanks for such a clear and helpful pull request!