The main change is that now there is only one main state in the
machine (aka project), not a separate copy per machine.

This commit also fixes leave-{blk,sub} events and adds several new
observations, such as exn-raised, that occurs every time a machine
switches to an exceptional control flow.

The commit also adds few common method to the monad interface.

Now it is possible to attach arbitrary values to a bitvector.

also fixes a bug in Eval.binop.

This bug leads to a type error, as a result of a byte load and a half
word load is stored in a 32 bit variable.

a type checker didn't held correctly shifts
and the normalizer didn't properly recurse

allow a backend to choose whether to propagate consts or not.

this is temporal cludge.

BIR becomes unreadable. We will apply normalization when needed.

Later, we may add a memoization or stuff like that.

this enables better diagnostics of compilation and configuration errors
on travis.

the plugin will mark all terms that were visited by Primus with the
visited attribute.

Should do this long ago, but finally found some time. No more
unnecessary parentheses!

Note: operator precedences follow the C language, not OCaml.

   +-------------------------------------------+----+
   | x[y], <cast>:<N>[x], extract:<N>:<M>[x]   | 10 |
   +-------------------------------------------+----+
   | ~x, -x                                    |  9 |
   +-------------------------------------------+----+
   | *,/,/$, %, %$                             |  8 |
   +-------------------------------------------+----+
   | +,-                                       |  7 |
   +-------------------------------------------+----+
   | <<,>>,~>>                                 |  6 |
   +-------------------------------------------+----+
   | <,>,<=,>=                                 |  5 |
   +-------------------------------------------+----+
   | =, <>                                     |  4 |
   +-------------------------------------------+----+
   | &                                         |  3 |
   +-------------------------------------------+----+
   | ^                                         |  2 |
   +-------------------------------------------+----+
   | |                                         |  1 |
   +-------------------------------------------+----+
   | let.., x with y <- z, if.., . (dot)       |  0 |
   +-------------------------------------------+----+

1) Higher precedence means that operators binds tighter.
2) Any bil statement has precedence lower than an expression.

1. at some branches the simplified didn't recurce
2. if a cast casts to the same type it is removed

it was wrong.

The ssa just translates a program into the SSA form
The dead-code-elimination is a conservative deadcode elimination
plugin, that helps alot with x86 binaries by removing tons of unused
flags calcualations.

The string_of_value function now emits the `0x` prefix for hexnumbers,
unless instructed explicitly with the `prefix:false` flag.

The reason for this change is to enable a consitent handling of signed
and unsinged, negative and positive values.

long time ago we forgot to make it public.

1. Now it will run until a fix point is reached
2. A simple constant propagation is added, so that more virtuals are
removed

there is no need to have two different names for memory.

1. The original implementation emited the following code

```
    flag := flag
```
   to denote an unchanged flag. The code was problematic for both static
   and dynamic analysis. Static analysis treated this as a used before
   defined variable and treated flag as a free variable in the eclosing
   definition. The dynamic analysis treated `flag` as an undefined
   variable with all the consequences.

   The new implementation uses the `if` statement and changes flag only
   if it is necessary. Moreover, instead of using multiple if/then/else
   expressions, all assignment are gathered under one big if statement,
   that leads to a more simple CFG representation.

2. The original implementation emited a temporary variable that holds a
   number of bits to which the destination should be shifted. There is
   no need for such variable (it is never changed). Moreover, the
   value of this variable is usually (if not always) a constant, so it
   is better to inline it directly in expressions. It looks like that
   this is the case of a common error -  delegating OCaml computation to
   BIL. In our case this problem is solved by constant propagation.

If in operator is associative and expression is left recursive, then
recurse to the right in hope to meet constants there. In particular,
this will simplify `x + 1 + 1 + 1` to `x + 3`.

1. do not store values with just empty sets of values
2. special handling for the one byte storage
3. remove a key from the taint set if it is set to empty

so that profiler will work, as well as other at exit handlers.

and rolls back the [Word.string_of_value] behavior.

The pretty printing function was totally rewritten from scratch, as
Z.format and other Z printing functions are broken. Now we have a
generic printer that suits all tastes, as well as 9 preinstantiated
pretty printers.

We also changed the bitvector default string representation. The new
representation doesn't used [true] or [false] for the [1] and [0] (this
is binary analysis anyway), and doesn't loose the signedness
information.

Updated bap.mli with a documentation for new functions.
Also, renamed Image.Scheme.reference to Image.Scheme.relocation.

Now it is enough just to type `make doc` and everything will work out of
box, if a correct version of OCaml is used, i.e., 4.03. The Makefile
recipe will pull the latest version of argot (git is required) and use
it to build the documentation. BAP should be installed either from the
source tree, or from opam, doesn't matter.

also removes some unnecessary stuff from it.

I've removed few unnecessary funciton, e.g., `State.modify` and also
fixed the type of the call/cc function.

also fixes few bugs that were found as a result

1) a zero width word was created (should be a one bit width)
2) a bug in the typechecker that led to a stackoverflow

So far no more type errors

also added parameters to the primus loader.

because my last update to primus broke it(((

The __primus_linker_unresolved_call function is called in case if a
linker can resolved a call.

Also adds the `pc` method to the interpreter interface and publish it
as a primitive `get-current-program-counter`.

not actually in the random number generator, but rather in a casting it
to a word.

just fail the machine.

Now we can specify a list of entry points, or a magic `all-subroutines`
to start execution in parallel from all specified entry points. If no
specified, then all subroutine terms marked with the `entry_point`
attribute are entered.

The primus-limit plugin will terminate a machine after a certain amount
of computations has happened.