Add JMesh sample files, parsers, fix timing functions, and update benchmark results

[fangq]

hi @neurolabusc, I added a few mesh files stored in the NeuroJSON/JMesh format, these files are

• zlib.jmsh: JSON/JMesh based mesh file; MeshVertex3 stored in single-precision; both MeshVertex3/MeshTri3 are binary data compressed with zlib(zip) compression, then base64 encoded
• zlib.bmsh : binary JSON (BJData) based JMesh file, like zlib.jmsh, except that data are not base64 encoded
• raw.min.json : minimized pure JSON file with array-shaped MeshVertex3 and MeshTri3
• raw.bmsh : same as zlib.bmsh, except that the binary node/face data are not compressed

I also found that the previous elapse time estimation approach (d.getMilliseconds) produces wildly varying results from each run. I changed it to d=Date.now() and Date.now() - d. This gives stable timing results, and also matches the timing based on console.time('mytimer') and console.timeEnd('mytimer')

I added two lightweight parsers

• lib/jdata.js to parse JData annotations in the JSON, and
• lib/bjdata.js to parse BJData files

Neither of these have been optimized. jdata.js converts an ND array to a numjs object, although I disabled the conversion, so the output are TypedArray 1D vectors

I also produced lzma (ALONE) based JMesh files, although I am not yet able to add the lzma inflation/deflation support in the parser. But you can see from below that the lzma.bmsh can be quite small.

-rw-rw-r-- 1 fangq fangq  2295259 Mar 27 22:34 lzma.bmsh
-rw-rw-r-- 1 fangq fangq  3102700 Mar 27 22:33 lzma.jmsh