Refactor docsystem internals.

[MichaelHatherly]

Refactor docsystem internals

The first commit consists of the following changes:

Structure of a module's #meta dict now has the following layout in all cases (with the exception of keywords):

• each module's #meta ObjectIdDict has keys of type Binding and values of type MultiDoc.
• each MultiDoc stores a collection of DocStr objects representing a single docstring each. These are ordered based on their order of definition rather than the current type_morespecific approach.
• DocStrs store the raw text of a docstring and lazily parse this to a Markdown.MD object when requested. By not parsing every docstring we also make some space savings in the sys.* files.

By keying #meta by Binding in every case the rest of the logic in doc! and doc becomes a lot more straightforward. at-doc expression building has also been simplified.

Additionally, the "summaries" displayed when no documentation can be found have been refactored and simplified.

Rewrite genstdlib.jl

The second commit takes advantage of the simplifications in the previous commit to reduce the complexity in this script. Outwardly the only difference should be the formatting of the summary that is displayed and speed (where timings drop from ~18 to ~9 seconds.)

[tkelman]

what is this in addition to?

[MichaelHatherly]

It was in addition to the default no docs message that appears, but it's not really necessary so I've remove the "Additionally".

[tkelman]

Aside from little nitpicks, this does look like a really nice simplification. The fewer things we have to implicitly import into base to make the docsystem work, the better, and 💯 for lazy parsing of the docstrings.

[MichaelHatherly]

Thanks for reviewing @tkelman, I've pushed fixes for those comments.

[MichaelHatherly]

Found and fixed an issue in parsedoc where the generated markdown did not include the same metadata as it used to, namely a :path and :module.

Also added a small third commit that returns some extra metadata from @doc that will be useful when generating external docs. I'm happy to make a separate PR later for this if that would be preferred.

[IainNZ]

Not sure the Americans will appreciate this name ;-)

[MichaelHatherly]

Yeah, probably not :( Will fix.

[MichaelHatherly]

Spellings changed in latest push.

[tkelman]

CI failures were #15294, restarted

From worker 4:       * numbers              Error During Test
    From worker 4:    Test threw an exception of type UndefVarError
    From worker 4:    Expression: round(T,true // false) === convert(T,Inf)
    From worker 4:    UndefVarError: no_op_err not defined
    From worker 4:     in gcd(::Bool, ::Bool) at ./intfuncs.jl:11
    From worker 4:     in Rational{Bool}(::Bool, ::Bool) at ./rational.jl:9
    From worker 4:     [inlined code] from /tmp/julia/share/julia/test/numbers.jl:2849
    From worker 4:     in anonymous at ./no file:4294967295
    From worker 4:     [inlined code] from ./essentials.jl:79
    From worker 4:     in include_string(::UTF8String, ::ASCIIString) at ./loading.jl:371
    From worker 4:     in include_from_node1(::ASCIIString) at ./loading.jl:420
    From worker 4:     [inlined code] from ./util.jl:179
    From worker 4:     in runtests(::ASCIIString) at /tmp/julia/share/julia/test/testdefs.jl:7
    From worker 4:     in (::Base.Serializer.__deserialized_types__.##9137)(::ASCIIString) at /tmp/julia/share/julia/test/runtests.jl:36
    From worker 4:     in run_work_thunk(::Base.##254#256{Base.CallMsg{:call_fetch}}, ::Bool) at ./multi.jl:714
    From worker 4:     [inlined code] from ./multi.jl:1010
    From worker 4:     in (::Base.##253#255{Base.CallMsg{:call_fetch},TCPSocket})() at ./task.jl:59

[MichaelHatherly]

Barring further comments, I'll merge this later on today.