We used to be storing packages from HTTP hosted sources in directories
named after the hostname. There was no distinction between a package
from an HTTP source and an HTTPS source with the same name. This was
contrary to the rest of pub's logic, which considered those different
sources.

Closes #1418

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//2028853002 .

If a path dependency had a dependency back on the root package, it
would be detected as always stale because the root package isn't in the
lockfile. This changes the check to always consider that fresh, since
all dependencies on the root package are satisfied.

Closes #1416

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1985063004 .

OfflineHostedSource was failing to parse the directory names for
prerelease package versions correctly.

Closes #1415

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1971523002 .

We no longer include overridden dependencies' constraints, since those
constraints aren't checked when doing version resolution. In the worst
case, including them caused us to generate invalid lockfiles with
"<empty>" as the SDK constraint.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1956613002 .

We had a logic error wherein we were checking dependencies only
for cached packages when we should have been checking them only
for *uncached* packages.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1956543002 .

This passes --format=ustar to the tar process on Linux and OS X, which
should prevent user-specific configuration from causing it to use the
PAX format which is incompatible with 7zip on Windows.

Closes #1411

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1920433002 .

R=nweiz@google.com

Review URL: https://codereview.chromium.org//1907023002 .

Closes #1405

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1892683002 .

This is necessary since the SDK's repo requires that.

R=kevmoo@google.com

Review URL: https://codereview.chromium.org//1902993003 .

BUG=https://github.com/dart-lang/sdk/issues/26295
BUG=https://github.com/dart-lang/pub/issues/1407
R=rnystrom@google.com

Review URL: https://codereview.chromium.org/1895423002 .

This was confusing browsers which ignore the ETag for XHRs and would
simply use the local cache for an asset even if it had changed on the
server.

R=sigmund@google.com

Review URL: https://codereview.chromium.org//1874623004 .

"git ls-files" doesn't natively resolve symlinks, so this adds some
logic to do so manually. This brings the behavior of uploading from a
Git repo in line with non-Git uploading.

We still don't have a multipart parser in Dart, so we unfortunately have
to way of writing automated tests that the tarballs are created as
expected. I did verify this behavior locally, though.

Closes #1400

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1862833002 .

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1863773003 .

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1852513004 .

Closes #1396

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1809993003 .