DTM: Leverage WARLness of hartsel to skip nonexistent harts

Resolves #268

I'm adding this to test DM clock gating, but it's a good idea in general.

This is an alternative to #53; see that PR for context.

Use $target_alias for TARGET_DIR if set

Define a chunked_memif_t interface that serves memif_t, and is implemented by htif_t and htif_hexwriter.  This allows the use of memif_t without needing to bring in htif_t.

This removes `bool permissive` from htif.cc, which was originally used
to control whether or not htif was parsing arguments permissively. This
instead uses getopt's builtin `int opterr` to hold this state. getopt
then uses this to print/not print to stderr on unknown options.

Resultingly, unknown VCS options will not produce any stderr output.

Signed-off-by: Schuyler Eldridge <schuyler.eldridge@ibm.com>

Can't spell good

Signed-off-by: Schuyler Eldridge <schuyler.eldridge@ibm.com>

This adds an option to disable permissive parsing (which can work as an
alternative to `--`) as `+permissive-off`.

This adds help text that HTIF can emit directly via a new `htif::usage`
method. This can be accessed from VCS, but not from
Verilotor/emulator.cc.

Signed-off-by: Schuyler Eldridge <schuyler.eldridge@ibm.com>

Signed-off-by: Schuyler Eldridge <schuyler.eldridge@ibm.com>

This adds an option `+permissive` that will enable old-style permissive
option parsing. Any option that HTIF does not understand it will ignore.
This must be explicitly turned off with `--` to indicate where the
binary is.

Signed-off-by: Schuyler Eldridge <schuyler.eldridge@ibm.com>

Approach for exposing HTIF options to other tools

This retains the legacy htif_t constructor using a vector of string to
pass command line arguments.

Signed-off-by: Schuyler Eldridge <schuyler.eldridge@gmail.com>

Does some minor cleanup of usage formatting.

Adds a htif_t() constructor that does not do argument parsing. This is
needed for something like `elf2hex` which wants to pass no arguments
to htif_t.

Note: `elf2hex` is currently non-working on master and not as a result
of this.

Signed-off-by: Schuyler Eldridge <schuyler.eldridge@gmail.com>

This changes the `htif_t` constructor to use getopt for option parsing
adding support for POSIX-like `--` long options while preserving
support for "legacy" `+` args. The options supported by `htif_t` are
exposed to programs importing the `htif.h` header.

This results in a change of the `htif_t` and `dtm_t` constructor APIs
so that it can directly process command line arguments. Previously
the `htif_t` constructor was:
  htif_t::htif_t(const std::vector<std::string>& target_args)
Now it is:
  htif_t::htif_t(int argc, char** argv)

As `htif_t` is a secondary consumer of command line arguments (e.g.,
it is constructed after the rocket-chip emulator), the options
supported by `htif_t` are exposed via three new macros in `htif_t.h`:
  * HTIF_USAGE_OPTIONS -- usage string suitable to be appended to any
    program that will pass arguments to `htif_t`
  * HTIF_LONG_OPTIONS_OPTIND -- Base val for `htif_t` options
  * HTIF_LONG_OPTIONS -- `struct option` fragment for `htif_t`. All
    options index starting from `HTIF_LONG_OPTIONS_OPTIND`.

Support for the currently unsupported option `+disk` now raises an
exception.

Space is provided for non-standard options which may be exposed to
`htif_t`. Currently only the VCS options of the form `-cm XXX` are
ignored.

This fixes the `syscall_t::set_chroot` method so that it will write
out the location that it failed to chroot to, `where,` as opposed to
the unset attribtue `chroot`.

dtm: Fix HALTREQ/RESUMEREQ issue and don't just die

dtm: Make sure to clear HALTREQ/RESUMEREQ when done. Don't die, just print an error, to allow getting debugging waveforms when badness happens.

* dtm: Start all harts at _start, not just hart 0

* DTM: additional cleanup for halt/resume different harts

* dtm: Proper enumeration

Whitespace fixes

Debug v013

debug_v13: Support Debug v13