Fix wrong package name in build instructions

This corresponds to the change in the install instructions for Ubuntu
outlined in #335.

Make bpftrace -l list sofware and hardware types (#44)

tracepoint shortcut t breaks args

hist: split negative, zero, and one into separate buckets

Add --version

- Also refactored code in attached_probe.cpp and
ast/semantic_analyzer.cpp to use the sw/hw lists instead of the
hardcoded probe names.

lhist: interval notation tweak

runqlat.bt: remove if semicolon

if semicolon

Link against system bcc

Previously, bpftrace was clone bcc from github at build time, building
a version of bcc, then linking against the built bcc. This is a no-go
for many packaging systems including Fedora, openSUSE, and the internal
Facebook one.

This patch requires bcc and bcc development headers be installed to
build bpftrace and removes the build time downloading behavior.
Also simplifies some of the header search logic.

Test plan:
```
$ ./build/tests/bpftrace_test
...
$ sudo make -C build runtime-tests
...

$ cat ~/scratch/bpftrace/path.bt

kprobe:vfs_open
{
    printf("open path: %s\n", str(((path *)arg0)->dentry->d_name.name));
}

$ sudo ./build/src/bpftrace ~/scratch/bpftrace/path.bt
[sudo] password for dlxu:
Attaching 1 probe...
open path: stat
open path: user-745673103.journal
open path: user-1001.journal
open path: user-745673103.journal
open path: user-1001.journal
open path: user-745673103.journal
open path: user-1001.journal
open path: iio:device0
open path: stat
...

`*-inl.h` files, or inline headers, are an implementation detail. Users
of an interface that is backed by an inline header should not concern
themselves with the details of implementation.

This patch normalizes the interface.

Avoid crashing if mistakenly just '-d' or '-v' is used

Currently, if by mistake just '-v' or '-d' is used without any
additional parameter (no filename nor -e 'program') bpftrace crashes:

./src/bpftrace -v
terminate called after throwing an instance of 'std::logic_error'
what():  basic_string::_M_construct null not valid
Aborted (core dumped)

This commit fix this problem.

Return cleanly after printing help

Test plan:
```
 $ ./build/src/bpftrace -h &> /dev/null && echo $?
 0
```

This allows operators to use other tools, like bpftool, to
introspect a running program.
Signed-off-by: David Calavera <david.calavera@gmail.com>

Fix use-after-free in BPFtrace::get_arg_values

Docs

implement review feedback: pass vector by const ref, use reinterpret cast, remove move() of returned string

we were moving as far as constructor arg, but afterwards we were copying to member. this fixes that. https://stackoverflow.com/a/16725030/5257399

implement review feedback: avoid copying strings, prefer make_unique, use standard naming convention for private members

Previously get_arg_values was returning a vector of uint64_t values
that could be passed directly to printf(3). For string values
get_arg_values was returning a pointer to a char*. For some cases it
was attempting to handle freeing the char* memory via a stack
allocated std::vector. Unfortunately, this was stack allocated in
get_arg_values so the char* data would get freed before it was used in
the subsequent call to printf().

In other cases get_arg_values was not freeing char* values and was
leaking memory (probe, stack, and ustack).

get_arg_values() now returns a vector of objects of type IPrintable
instead of uint64_t values. Each object has a method .value() that
returns the uint64_t value usable by printf(). For strings this allows
us to keep around the original std::string until after we've called
printf(), so we don't need to strdup() anymore.

Fixes #194

improve tracepoint not found message

[test] fix runtime-tests cmake settings