Allow for long Http* error.log lines

[bneradt]

Many URLs and log paths are long enough that the previous 256 byte
error.log limit for HttpSM and HttpTransact made it so that the logs got
truncated. Some of those logs had meaningful strings at the end which
made them hard to interpret unless you had the source code in front of
you to figure out what was truncated. This utilizes bwprint and local
std::string buffers for arbitrarily long log lines.

[bneradt]

[approve ci autest]

[bryancall]

Thread local was actually slower in the benchmark I did.

Benchmark results:

----------------------------------------------------------------------------
Benchmark                                  Time             CPU   Iterations
----------------------------------------------------------------------------
thread_local_benchmark/threads:12        139 ns         1496 ns       538452
local_scope_benchmark/threads:12        6.68 ns         80.0 ns      8045064

Code:

#include <benchmark/benchmark.h>
#include <thread>

//----------------------------------------------------------------------------
static void thread_local_benchmark(benchmark::State& state) {
  for (auto _ : state) {
    thread_local std::string str;
    str.append("dlk alsdkfjsaldkfj alskdfjasldk alskdfjlasdk ldsasdkfjalsdkfj lkj falskdjfla sdjfjlsjdfl alsdjkflsfj");
    str.append("jfghkjirtyu fj alskdfjasldk alskdfjladk lasdkfjalsdkfj lkj falskdjfla sdjfjlsjdfl alsdjkflsfjdlk alsdkfjsaldkasdfasdfasd asdfasddf asdfasdf asdfasdf asdfasdfasdf afghjfghk alskdfjlasdk lasdkfjalsdkfj lkj falskdjfla sdjfjlsjdfl alsdjkflsfjdlk alsdkfjsaldkasdfasdfasd asdfasddf asdfasdf asdfasdf asdfasdfasdf afghjfghkjirtyu fj alskdfjasldk alskdfjlasdk lasdkfjalsdkfj lkj falskdjfla sdjfjlsjdfl alsdjkflsfjdfasd asdfasddf asdfas fj alskdfjasllasdk lasdkfjalsdkfj lkj falskdjfla sdjfjlsjdfl alsdjkflsfj");
  }
}

//----------------------------------------------------------------------------
static void local_scope_benchmark(benchmark::State& state) {
  for (auto _ : state) {
    std::string str;
    str.append("dlk alsdkfjsaldkfj alskdfjasldk alskdfjlasdk ldsasdkfjalsdkfj lkj falskdjfla sdjfjlsjdfl alsdjkflsfj");
    str.append("jfghkjirtyu fj alskdfjasldk alskdfjladk lasdkfjalsdkfj lkj falskdjfla sdjfjlsjdfl alsdjkflsfjdlk alsdkfjsaldkasdfasdfasd asdfasddf asdfasdf asdfasdf asdfasdfasdf afghjfghk alskdfjlasdk lasdkfjalsdkfj lkj falskdjfla sdjfjlsjdfl alsdjkflsfjdlk alsdkfjsaldkasdfasdfasd asdfasddf asdfasdf asdfasdf asdfasdfasdf afghjfghkjirtyu fj alskdfjasldk alskdfjlasdk lasdkfjalsdkfj lkj falskdjfla sdjfjlsjdfl alsdjkflsfjdfasd asdfasddf asdfas fj alskdfjasllasdk lasdkfjalsdkfj lkj falskdjfla sdjfjlsjdfl alsdjkflsfj");
  }
}


BENCHMARK(thread_local_benchmark)->Threads(12);
BENCHMARK(local_scope_benchmark)->Threads(12);
BENCHMARK_MAIN();

[bneradt]

Thread local was actually slower in the benchmark I did.

Thank you for doing the benchmark! I'll use an undecorated std::string. Either way, this isn't hot code. Might as well keep it more simple, especially if thread_local can pessimize performance.

[bryancall]

@cmcfarlen will review this

[cmcfarlen]

We talked about moving this libfmt style print into Log itself to support this directly, but this is a bit more work than intended for this PR. I'm going to make an issue to add better support for this if someone wants to work on it in the future.

[bryancall]

Based on @cmcfarlen review

[zwoop]

Cherry-picked to v9.2.x

[zwoop]

Cherry-picked to v9.2.x