Fixes: #3642

We need to ensure that vmods use the same functions and data
types as varnish-cache.

Also it seems that having the define in autoconf might suggest
that it could be changed, which would be misleading.

For now, this is only VRE_unpack() to get the pcre2_code handle.

If we ever change the regex backend again, we will remove
vre_pcre2.h and add something else.

Suggested by Dridi, thus authorship attributed to Dridi.

Refs 31e3895a

Adapted to the pcre2 change, and (manually) tested both with and
without jit compilation and various tweaks to trigger both match
and depth limits.

Closes #3658

Before 6014912e, VRE_sub() used an
ovector of size 30, which always containted sufficient space to store
the 10 possible back- references \0 thorugh \9.

Now that we use pcre2_match_data_create_from_pattern() and later
pcre2_get_ovector_pointer(), we only get space for the number of
substrings in the pattern, see pcre2api(3):

	The ovector is created to be exactly the right size to hold
	all the substrings a pattern might capture.

Consequently, we need to check that back references do not exceed the
maximum ovector.

Spotted by: Coverity

When receiving H/2 data frames, make sure to take the advertised content
length into account, and fail appropriately if the combined sum of the
data frames does not match the content length.

The change to VTCP_Check() in 58a21da7
broke expect_close in varnishtest.

In RST, there needs to be an empty line after ::, otherwise the whole
block is interpreted as a normal paragraph where newlines are not
preserved.

Now that VRE is the only regular expression API we use, we can migrate
its backend to pcre2. The existing 'pcre_*' parameters are also renamed
to reflect this migration, and 'pcre_match_limit_recursion' gets special
treatment and is renamed to pcre2_depth_limit.

This creates an additional API breakage in VRE: the `match_recursion`
field in `struct vre_limits` is also renamed. One last breakage is the
removal of `VRE_has_jit` used by only one undocumented varnishtest
feature, and the pcre_jit feature is only used by one test case that no
longer fails.

The pcre jit compilation feature was broken anyway: sealing it at
compile time will not reflect what VRE actually links to. Once we have
a test case needing the jit feature, we can introduce a better API for
that check.

There is one outstanding performance problem, the ovector that was
previously allocated on the stack now needs to be allocated from the
heap. It might be possible to implement a pcre2 context to fix that or
maybe pool them, but for now we have heap allocations on the critical
path. The VRE_sub() function makes sure to make a single ovector
allocation (technically a pcre2_match_data allocation) since it's the
only one guaranteed to loop on a single regular expression for the
`regsuball()` use case.

On the documentation front, the SmartOS installation instructions are
hidden for lack of a pcre2 package.

Closes #3616
Closes #3559

This reverts commit f254fff7.

Conflicts:
	.circleci/config.yml

Since we turn warnings into errors, that means failing the build because
something we have no control over in /usr/include does not have a strict
prototype or some other shenanigan.

Just in case it might be added by Wall or Wextra in a future clang or
GCC release, we may disable it explicitly too.

Refs #3565

Bans using regular expressions will consume slightly more space, but
more importantly that breaks persistence binary compatibility. That's
not a concern because we are both planning for a major release where
that kind of breakage is acceptable, and in the context of a pcre2
migration we would also break ban persistence.

And now, VRE is the sole pcre consumer.

It packs a vre_t and a pcre in a single allocation that can be used by
both VRE_match() and VRE_sub().

We always use it internally except:

- where the pcre2_jit_compilation parameter applies
- when libvcc verifies that a regular expression compiles

For the latter, the verification would attempt a jit compilation as well
until now, but we no longer need to waste cycles on that.

This is a major step back in terms of error reporting, but I haven't
found anything in libpcre to translate an error code to an error
message. "pcre error %d" should still be a good enough hint.

Looking forward, we are going to need this for libpcre2 but again the
API leaves some to be desired and only works by writing error messages
to buffers.

The return value implies that we got a valid error code, which will be
verified with pcre2. I couldn't find how to verify error codes with pcre.
Writing to the VSB is however fail-safe, it's the caller's job to deal
with a VSB error.

With the introduction of VRE_sub() many VRE_exec() use cases went away:

- VRE_NOTEMPTY flag
- start offset
- ovector for capture groups

The subject string length is now optional and zero can be passed to signal
that the subject string is null-terminated.

There are no options for VRE_match() but an option argument is kept in
case we start exposing some in the future.

This gives us a clean separation of VCL and pcre interactions.

Its first action ${string,repeat,<uint>,<string>} helps simplify many
unwieldy test cases that will hopefully be easier to edit from now on.

Instead of having a mere value, these would be able to compute a macro
expansion. We parse the contents inside the ${...} delimiters as a VAV,
but there can't be (yet?) nested curly {braces}, even quoted.

The first argument inside the delimiters is the macro name, and other
VAV arguments are treated as arguments to the macro's function.

For example ${foo,bar,baz} would call the a macro "foo"'s function with
arguments "bar" and "baz". Simple macros don't take arguments and work
as usual.

The latter operates on a VSB, which is always what call sites are doing
anyway. It also takes the responsibility of ignoring unknown macros, in
preparation for more responsibilities that will also require the ability
to fail a test case.