These functions expect that file_path can have base_path stripped from
it, but during reloading they can be given relative paths.  Maybe this
behaviour varies between the notify backends?

This fixes two zola serve panics on FreeBSD (poll backend).

Closes #756

* Set up CI with Azure Pipelines

[skip ci]

* Remove other CI + remove newline in anchor link tpl

* Some more Windows specific changes

Closes #726

Part fix of #742

Clippy is returning some warnings.  Let's fix or explicitly ignore
them.  In particular:

- In `components/imageproc/src/lib.rs`, we implement `Hash` explicitly
  but derive `PartialEq`.  We need to maintain the property that two
  keys being equal implies the hashes of those two keys are equal.
  Our `Hash` implementations preserve this, so we'll explicitly ignore
  the warnings.

- In `components/site/src/lib.rs`, we were calling `.into()` on some
  values that are already of the correct type.

- In `components/site/src/lib.rs`, we were using `.map(|x| *x)` in
  iterator chains to remove a level of indirection; we can instead say
  `.copied()` (introduced in Rust v1.36) or `.cloned()`.  Using
  `.copied` here is better from a type-checking point of view, but
  we'll use `.cloned` for now as Rust v1.36 was only recently
  released.

- In `components/templates/src/filters.rs` and
  `components/utils/src/site.rs`, we were taking `HashMap`s as
  function arguments but not generically accepting alternate `Hasher`
  implementations.

- In `src/cmd/check.rs`, we use `env::current_dir()` as a default
  value, but our use of `unwrap_or` meant that we would always
  retrieve the current directory even when not needed.

- In `components/errors/src/lib.rs`, we can use `if let` rather than
  `match`.

- In `components/library/src/content/page.rs`, we can collapse a
  nested conditional into `else if let ...`.

- In `components/library/src/sorting.rs`, a function takes `&&Page`
  arguments.  Clippy warns about this for efficiency reasons, but
  we're doing it here to match a particular sorting API, so we'll
  explicitly ignore the warning.

* Add hard_link_static config option.

* Copy or hardlink file depending on an argument.

Modify the call sites for `copy_file` to account for the extra argument.

* Plug the config setting through to copy_file.

Don't apply the config option to theme's static directory.

* Update documentation.

* Backticks make no sense in this comment.

* Addressing PR comments.

* Be consistent with argument naming.

Closes #722

Signed-off-by: Igor Gnatenko <i.gnatenko.brain@gmail.com>

Without this, at least one search engine spider was generating a
validation warning.

* chore: Update glob to 0.3

Signed-off-by: Igor Gnatenko <i.gnatenko.brain@gmail.com>

* chore: Update ws to 0.8

Signed-off-by: Igor Gnatenko <i.gnatenko.brain@gmail.com>

Close #693

Close #686

* Add check subcommand

* Add some brief documentation for the check subcommand

* Start working on parallel link checks

* Check all external links in Site

* Return *all* dead links in site

Justification for this feature is added in the docs.

Precedent for the precise syntax: Hugo.

Hugo puts this syntax behind a preference named headerIds, and automatic
header ID generation behind a preference named autoHeaderIds, with both
enabled by default. I have not implemented a switch to disable this.

My suggestion for a workaround for the improbable case of desiring a
literal “{#…}” at the end of a header is to replace `}` with `}`.

The algorithm I have used is not identical to [that
which Hugo uses][0], because Hugo’s looks to work at the source level,
whereas here we work at the pulldown-cmark event level, which is
generally more sane, but potentially limiting for extremely esoteric
IDs.

Practical differences in implementation from Hugo (based purely on
reading [blackfriday’s implementation][0], not actually trying it):

- I believe Hugo would treat `# Foo {#*bar*}` as a heading with text
  “Foo” and ID `*bar*`, since it is working at the source level; whereas
  this code turns it into a heading with HTML `Foo {#<em>bar</em>}`, as
  it works at the pulldown-cmark event level and doesn’t go out of its
  way to make that work (I’m not familiar with pulldown-cmark, but I get
  the impression that you could make it work Hugo’s way on this point).
  The difference should be negligible: only *very* esoteric hashes would
  include magic Markdown characters.

- Hugo will automatically generate an ID for `{#}`, whereas what I’ve
  coded here will yield a blank ID instead (which feels more correct to
  me—`None` versus `Some("")`, and all that).

In practice the results should be identical.

Fixes #433.

[0]: a477dd1646/block.go (L218-L234)

To also process the ones found in templates

Closes #669

Closes #674

This reverts commit 1815155c1d.

This feature was originally reported in #257 and got fixed in
3a2dab59743e8d193a80ddb1aa1932e1b5462322, however it seems this
got lost during development.

Closes #661
Ref #257

Close #633

Split sitemap when it is getting too big