spreadsheet: speedup by removing intermediate copies

spreadsheet: speedup by removing intermediate copies

The passed wchar_t* array should only be copied once before working on
it. During the whole processing, there should be no copy on the common
case (but extra copies might be needed for corner cases). Performance
improvement is also highly related to wchar_t* Scilab string storage.

This commit patch csvTextScan and csvRead functionnalities; please note
that using mgetl and csvTextScan will still need 2 extra copies than
csvRead (1 to store mgetl results and 1 to have mutable buffers during
scanning).

Change-Id: I51f8c4b0e0437bfe4ac69439b739a60262da1cf1