Optimize ctags calls

It's expensive to call ctags once per source file identified in the
DWARF data.  This patch:

- Calls ctags on all the source files from a single ELF at once. '-L -'
  is used to avoid risk of hitting OS limits with a huge command line.

- Similarly, reads the output a line at a time to avoid holding a huge
  amount of ctags output at once in memory.

- Recovers the file paths from the lines output by ctags, which it
  always prints in '-x' mode, even for a single input file. (Note that
  this will break on paths with embedded spaces.  This limitation was
  present in the original through the use of shell=True, so it is
  assumed to be acceptable here - consider json output if it becomes a
  problem).

- Renames 'workspace' to 'local_workspace' in FunctionLineNumbers for
  consistency with the rest of the program.  (For programming
  convenience, the cache of results is changed to key by local file name
  - i.e. with local_workspace prepended - rather than the caller's file
  name, but is otherwise unchanged).

Note the apparent loss of specificity in the error message for ctags
failure, which no longer outputs each file name.  This was probably
intended to report problems with particular missing, corrupted, or
unreadable files.  In practice, Exuberant ctags returns success in these
cases, even for a single file on the command line.  The only way I found
to provoke a nonzero exit code was to pass a bad command line argument.
The point is that ctags will either fail for every file or none of them,
so testing each exit code is not effective.  If file-specific errors
need to be reported, consider capturing and logging lines of stderr.

Change-Id: I89c80a8ac181d6bc01351c0f4922941829a1ed5c