This commit enhances the license detection that we have for various
licenses. Here we improve the detection for all licenses (especially the
Apache 2.0 License), the BSD 2-clause license, the MIT (with
attribution) license, and we add detection for the BSD 3-clause
license. One way that we achieved this improvement is by changing how
the license files are read so that rather than reading them as a
multi-line string which ended up represented as "[line1, line2, line3,
...]" internally, we read the full bytes of the license text and replace
all whitespace with a single space so the license text is now loaded as
"line1 line2 line3". For the MIT license we add the actual license text
and remove the "MIT" string as not all copies of the license clearly
indicate that the text is the MIT license. We take a similar strategy
for the BSD-2 and BSD-3 clause licenses. With this change, we reduce the
number of "custom" licenses in the codebase from 31 to 2. The two
remaining appear to be truly custom licenses, not carrying licenses
identifiable by SPDX. A follow-up will address "unknown" licenses.
e467e67fd4