This change makes it so umbrella projects get ordered in a way more
in-line with what we do with non-umbrellas.

For example, a non-umbrella app (rebar3 itself) gives output like:

    └─ rebar─3.8.0+build.4270.refaece1d7 (project app)
       ├─ bbmustache─1.6.0 (hex package)
       ├─ certifi─2.3.1 (hex package)
       ├─ cf─0.2.2 (hex package)
       ├─ cth_readable─1.4.2 (hex package)
       ├─ erlware_commons─1.3.1 (hex package)
       ├─ eunit_formatters─0.5.0 (hex package)
       ├─ getopt─1.0.1 (hex package)
       ├─ hex_core─0.4.0 (hex package)
       ├─ parse_trans─3.3.0 (hex package)
       ├─ providers─1.7.0 (hex package)
       ├─ relx─3.28.0 (hex package)
       └─ ssl_verify_fun─1.1.3 (hex package)

The current (prior to this patch) output for an umbrella project looks
like this:

    ├─ bootstrap─0.0.2 (git repo)
    ├─ dirmon─0.1.0 (project app)
    ├─ fake─0.1.0 (project app)
    ├─ file_monitor─0.1 (git repo)
    ├─ peeranha─0.1.0 (git repo)
    │  ├─ gproc─0.5 (git repo)
    │  ├─ interclock─0.1.2 (git repo)
    │  │  ├─ bitcask─1.7.0 (git repo)
    │  │  │  └─ lager─2.1.1 (hex package)
    │  │  │     └─ goldrush─0.1.6 (hex package)
    │  │  └─ itc─1.0.0 (git repo)
    │  └─ merklet─1.0.0 (git repo)
    ├─ recon─2.2.2 (git repo)
    └─ uuid─1.5.0 (git repo)
       └─ quickrand─1.5.0 (git repo)

Which kind of looks odd because the layout represents the dependencies
of deps as declared by the rebar3 config files, but mix up the project
apps as dependencies. This turns out to break expectations, and some
external tools (such as Blackduck's dep detection -- see
d8a0f2281a/hub-detect/src/main/groovy/com/blackducksoftware/integration/hub/detect/detector/hex/Rebar3TreeParser.java)
which were developed with regular apps in mind, but not umbrellas.

This PR changes things by reordering all project apps at the root level
and putting them first, and then indenting all deps below them. The same
app now looks like:

    ├─ dirmon─0.1.0 (project app)
    └─ fake─0.1.0 (project app)
       ├─ bootstrap─0.0.2 (git repo)
       ├─ file_monitor─0.1 (git repo)
       ├─ peeranha─0.1.0 (git repo)
       │  ├─ gproc─0.5 (git repo)
       │  ├─ interclock─0.1.2 (git repo)
       │  │  ├─ bitcask─1.7.0 (git repo)
       │  │  │  └─ lager─2.1.1 (hex package)
       │  │  │     └─ goldrush─0.1.6 (hex package)
       │  │  └─ itc─1.0.0 (git repo)
       │  └─ merklet─1.0.0 (git repo)
       ├─ recon─2.2.2 (git repo)
       └─ uuid─1.5.0 (git repo)
          └─ quickrand─1.5.0 (git repo)

Now the project apps are back at the top, and all the dependencies
follow at one level of indentation below.

There is no recorded change for single-app projects.

Clear _build/bootstrap on each bootstrap run

update template moving maintainers into extras field

fix case clause when fetching hg dep

This should avoid common problems when upgrading across versions where
old code kind of messes up a lot of things and clearing _build keeps
being required.

update relx to 3.28.0

Remove unnecessary clauses in update_package/3

Update hex_core and add mirror_of repo config

- removed deprecated maintainers field from config template in
  rebar_prv_packages

Add --generator option for eunit

Add a --fail_fast switch to common test

bump erlware_commons to 1.3.1

re-expose download/3 for backwards compatibility in git deps related plugins

This switch aborts the rebar3 run after the first test failure is
reported.

Since Common Test does not support this feature and it can be useful for
interactive development (a demand from one of the teams I work with), I
decided to add the feature as an experimental CT hook.

The hook notices any test failure, and forcefully aborts the current
script execution, but only during the initialization phase of the next
test; this ensure that all other hooks for a post-test failure have the
time to do things right.

The CT logs _will_ be interupted, and hooks of all kinds may also
suffer.

Since this might be a bit tricky to support internally, I will be fine
with a review that results in "no, Fred. Please test this in a project
plugin first" or if I get told to use profiles to mature the feature
(i.e. `rebar3 as failfast ct ...`) with a custom hook declaration.

For issue https://github.com/erlang/rebar3/issues/1974
Revert "add rebar.lock to template gitignore"

This reverts commit 8fb8a91130.

This reverts commit 8fb8a91130.

For issue https://github.com/erlang/rebar3/issues/1974

- fixes systests on darwin where previously the recursive cp would fail
   due to everything being copied after the last forward slash.
   Specifically, everything in  `systest/all_SUITE_data/` would be
   copied but not `all_suite_data` dir itself which the tests expected.

Fixing systest to use up to date plugin

prevents some failures when running

implement mib_first_files support in rebar_compiler_mib

mostly copy/paste from rebar_compiler_erl

consider A-MIB imports from B-MIB:

rebar compile MUST FAIL if mib_first_files is ["mibs/A-MIB.mib"]
rebar compile MUST SUCCEED if mib_first_files is ["mibs/B-MIB.mib"]