Consistent handling of Encoding

[AshesITR]

Currently, we use readLines() when linting, which assumes the system encoding when parsing files.

This results in uncovered lines w.r.t. fixing column numbers of UTF-8 Multibyte characters (at least on Windows, cf. #739) and in bugs when the file encoding doesn't match the system encoding (cf. #541)

lintr should

• get a new setting encoding with a smart default (see below) overridable in the usual ways.
• respect this setting in get_source_expressions()
• Update the tests to force UTF-8 encoding where necessary

The smart default for encoding should pick the first valid option of this list:

1. The Encoding field in a package DESCRIPTION
2. The Encoding field from the .Rproj file in a project directory
3. UTF-8

This default is system-independent and thus should be safe to use.