The new algorithm improves efficiency with iteration in place of
recursion, and in a number of other ways:

- Avoid searches into dead-end branches. The traversal of all branches
was done because of the overlapping prefix case -- "foo" and "foobar"
both in the set. Now we just search the tree for a match, but before
descending into the next branch, check if there are other branches at
which the current prefix matches a terminating node.

- Only do string comparisons when we hit a terminating node.

- Mark terminating nodes with a flag in the tree, so that we don't go
looking for the null byte in the strings table during the search.

While we're here, rename the flag for the nibble search as hinib --
non-zero if and only if we inspect the most significant nibble at that
node. Also remove some dead code from QP_Insert().

May be advantageous for loop/branch prediction.

Theoretically, this reduces the probability of collisions. Benchmarks
don't show much of a difference.

This adds the .compile() method to set objects, required for the
use of .match().

Docs for the .compile() method are currently incomplete.

strlen() is also cheap if it has a SIMD implementation, so we can
afford this optimization to reject some strings immediately.

Cannot be used for prefix matches.

For "quadbit patricia tries", inspired by the work of Tony Finch:
https://dotat.at/prog/qp/README.html

Radix 16 tries, examining a nibble at a time, to make the tries
smaller and reduce pointer chasing.

Only call strcmp() once, when a node is reached that must be either a
hit or a miss.

The search may have matched a string that is actually a prefix of
the subject string, if a longer string with the same prefix is also
in the set.

This "happens" to give correct results for match(), but which()
would return the wrong value.

The fix uses strcmp() instead of memcmp(), but that is also
vectorized, where the C library uses vector instructions.

Vector extensions are common hardware now, as are C libraries that
use vector instructions to implement functions like memcmp(). So
we hand off compares to the lib to get the advantage.

For the same reason, we can afford to call strlen() on the subject
string to locate the terminating null, rather than scan for it.

Also, the match function descends through the trie to find a
potential match, and does the comparison only then, as is common
for trie/critbit/patricia implementations.

since varnish-cache ecef48518f3b3f4bbf28256e090bdbb5cd2b163c backends
can be NULL (as defined with backend <name> None)

Fixes #1

configure checks if you have lcov & genhtml; these can be specified
with --with-lcov and/or --with-genhtml.

If they are available, then make coverage does the following:

- make clean, then make check with CC=gcc and CFLAGS set so that
  inputs for gcov/lcov are generated.

- lcov creates the src/coverage subdir and generates a targetfile
  there.

- genhtml generates HTML reports in src/coverage.

See also https://github.com/varnishcache/varnish-cache/issues/2967
for a general discussion regarding an inconsistency in the WS_Reserve()
interface