Use various USDT probes in Java and Ruby, and `malloc()`
uprobe in C, to summarize object allocations by type or
by size. Print statistics at the end or at intervals, with
optional trimming of top types by number of instances
allocated or number of bytes.

Trace `pthread_start` calls and get the symbol for the
newly-started thread for printing. Also allow no-Java runs
with no language specified for pthread-only tracing.

`uthreads` attaches to thread start and stop USDT probes
and traces out thread creation and destruction events.
Currently, only Java is supported.

Add `-S` switch for tracing syscalls with or without high-
level language method calls. When `-S` is provided, attach
to all `sys_` and `SyS_` entry points in the kernel and
trace the number of calls and execution latency (if `-L`
was also specified).

This is now also a useful general-purpose tool for syscall
count/latency summarizing when a language is not specified.

Add Node support for gc__start and gc__done probes, which
includes GC type flag parsing.

Skeletons for method flow, object allocations, general
event statistics, and thread statistics.

For runtimes that make it available, we now extract GC
event information such as heap sizes and print this
information in addition to the GC duration and timestamp.

`ugc` monitors GC start and end events in Java, Python,
and Ruby, printing out a message when a GC cycle completes
along with the duration of the GC and an optional
additional description that depends on the target language.

NOTE: This WIP does not have any interesting additional
descriptions yet.

Usage example:

```
ugc java 1448	# monitor Java GCs in process 1448
```

`ucalls` is a new tool that uses the USDT probes in high-
level languages (Java, Python, Ruby) to summarize the
number of method calls and optionally their latency. This
is similar to `funccount` and `funclatency` but for methods
in Java, Python, Ruby, and other languages that might have
USDT probes for method entry and exit.

`ucalls` uses the `method__entry` and `method__return`
probes in Java and Ruby, and the `function__entry` and
`function__return` probes in Python. The Python probes are
not as accurate because they do not report the class name,
but are still useful.

The optional `-L` switch indicates whether latency numbers
are required. By default, only the number of calls to each
method is printed. The `-T` switch filters out the top
methods (by number of calls or total latency).

Add USDT support to C++ API

Improve linear histogram limit, and improve error message.

runqlen: use bpf_get_smp_processor_id()

add runqlen tool

check effective userid on run #850

add Quick Start Guide for bcc docker

docs: Update eBPF features list

Update of BPF features list, following the release of kernel 4.9.

offwaketime: update with get_stackid fixes #437

Expose perf event support in C++ API

Improve C++ API perf buffer polling

add support for bpf map flags

Make sure program FDs get cleaned up correctly

Fix for issue 829 I reported 

Needed to make array initialized to null before bpf_probe_read as the
verifier was rejecting the original code. will probably need to fix the
tutorial as well

updating my local repo