When running under testnode, there was sometimes this failure:

```
File "/srv/slapgrid/slappart8/srv/testnode/bee/inst/test0-0/parts/slapos.recipe.cmmi/slapos/recipe/cmmi/README.txt", line 521, in README.txt
Failed example:
    print(system(buildout))
Expected:
    Uninstalling package.
    Installing package.
    package: [ENV] CFLAGS = -I/sw/include
    package: [ENV] LDFLAGS = -L/sw/lib -L/some/extra/lib
    package: [ENV] TMP = /sample_buildout/parts/package/tmp
    package: Applying patches
    patching file configure
    patching file Makefile.dist
    patched-configure --prefix=/somewhere/else --with-threads --without-foobar
    building patched package
    installing patched package
    installing patched package-lib
    <BLANKLINE>
Got:
    Uninstalling package.
    Installing package.
    package: [ENV] CFLAGS = -I/sw/include
    package: [ENV] LDFLAGS = -L/sw/lib -L/some/extra/lib
    package: [ENV] TMP = /sample_buildout/parts/package/tmp
    package: Applying patches
    patching file configure
    patching file Makefile.dist
    patched-configure --prefix=/somewhere/else --with-threads --without-foobar
    building patched package
    installing patched package-lib
    installing patched package
    <BLANKLINE>
```

The order of in which `installing patched package-lib` and ` installing patched package` are printed is not always same.

This test does `make install install-lib` with a Makefile containing:

```
all:
        @echo building package

install:
        @echo installing package

install-lib:
        @echo installing package-lib
```

@vpelletier  and I investigated this and he pointed out that order of parts depends is not stable when make is invoked with `-j` option.

Turned out that [on test node](https://nexedi.erp5.net/test_result_module/20180308-55459479/7) `MAKEFLAGS` is  set to `-j12`.


/reviewed-on !5

option is used, so this warning is incorrect.

There is no `Unused options for package: 'md5sum'.` (and this option is
used, so we don't want to ignore this)

Instead of using ellipsys use buildout provided way of ignoring these
messages.

Use buildout testing utility to start a webserver instead of downloading
from the internet.
This also fixes the test because the downloaded content changed.

  Update the file to reflect current implementation, as the tests
  were left w/o run for very long, some messages outputs were changed
  overtime.

  This may cause an import chicken-egg problem, so this also reduce
  the amount of imports to be done.

This reverts commit 18402f9f.

This reverts commit 3e54ec0c.

Use slapos.extension.strip instead.

This reverts commit f26567c4.

  And it is also backward compatible with buildout<2 and with current tests

It looks even better to change the default value on Download.__init__()

And something should also be done not to cache something local.