* Attempting to fix AppVeyor builds

* Trying again, 2015 image this time

* Use Appveyor's Go 1.11 stack

* Restore GOPATH\bin to PATH and delete old image config

* Add gcc to path manually

Default fastcgi timeout is 60 seconds
Add tests

quic-go now vendors all of its dependencies, so we don't need to vendor
them here.

Created by running:
gvt delete github.com/lucas-clemente/quic-go
gvt delete github.com/bifurcation/mint
gvt delete github.com/lucas-clemente/aes12
gvt delete github.com/lucas-clemente/fnv128a
gvt delete github.com/lucas-clemente/quic-go-certificates
gvt delete github.com/aead/chacha20
gvt delete github.com/hashicorp/golang-lru
gvt fetch -tag v0.10.0-no-integrationtests github.com/lucas-clemente/quic-go

Adding TLS client cert placeholders #2217 uses features of go
v1.10.  Update README requirements accordingly.

Update paths for filebrowser plugins

* Fix 502 errors for requests without headers

* Add unexported roundRobinPolicier

We have to preserve state for fallback mode of Header policy, so
it's required to save state in some variable

* Adding TLS client cert placeholders

* Use function to get the peer certificate

* Changing SHA1 to SHA256

* Use UTC instead of GMT

* Adding tests

* Adding getters for Protocol and Cipher

This makes setting the flag easier in projects that embed package caddymain.

Current Caddy code used a combination of CSS styles that
some mainstream browsers (e.g. Firefox) do not support well:
"td:first-child { width: 100%; }" together with
"td:last-child { padding-right: 5%; }".

The old approach was three columns with:
- "Name": 100% width, 5% padding left
- "Size": minimal width
- "Modified": minimal width, 5% padding right

Now the new approach is five columns with:
- <Dummy>: 5% width
- "Name": 80% width
- "Size": minimal width
- "Modified": minimal width
- <Dummy>: 5% width

Browser defaults vary and this way we prevent mix-ups
from setting only some colors.

* Fix a few import problems: snippets and import literals.

Two problems are fixed by this code simplification:
1. Snippets defined in one import file are strangely not available in
   another.
2. If an imported file had a directive with an argument "import", then
   the rest of the tokens on the line would be converted to absolute
   filepaths.

An example of #2 would be the following directive in an imported file:
    basicauth / import secret

In this case, the password would actually be an absolute path to the
file 'secret' (whether or not it exists) in the directory of the imported
Caddyfile.

The problem was the blind token processing to fix import paths in the
imported tokens without considering the context of the 'import' token.

My first inclination was to just add more context (detect 'import' tokens
at the beginning of lines and check the value tokens against defined
snippets), however I eventually realized that we already do all of this
in the parser, so the code was redundant. Instead we just use the current
token's File property when importing. This works fine with imported tokens
since they already have the absolute path to the imported file!

Fixes #2204

* renamed file2 -> fileName

* Fix copy/pasted comment in test.

* Change gzip example to basicauth example.

This makes it more clear how the import side effect is detrimental.

This fixes the permissions on /etc/caddy to match standard linux
permissions for /etc, and makes the Caddyfile read-only for the caddy
user.

This adds new feature to load envs from file provided from command line argument
Implement parsing of the env file for simple KEY=VALUE format

Add REQUEST_SCHEME to fastcgi envs

Fixes https://github.com/mholt/caddy/issues/2152

fastcgi: strip PATH_INFO from SCRIPT_FILENAME (mirroring SCRIPT_NAME)

Otherwise it overflows int type on 32-bit builds

This way we store a short 8-byte hash of the UA instead of the full
string; exactly the same way we store TLS ClientHello info.

And fix a typo in a comment, sigh

Knowing whether Caddy is running in a container is super-useful for
debugging and troubleshooting, as well as for making development-time
decisions, because Docker is one of the top contributors to our
user support burden.

Thanks to Eldin for helping to test it.