Document that hexfloat 'a'/'A' always emits the 0x prefix

[ny000815]

Problem

The documentation for the a/A (hexfloat) type in doc/syntax.md is inaccurate in two ways:

1. The example is shown as 1.8p+1, without the 0x prefix.
2. It states that the # flag "adds a 0x prefix".

Both are wrong. In practice fmt always emits the 0x prefix (matching printf's %a), regardless of #.

Reproduction

#include <fmt/format.h>
#include 
#include 
int main() {
  fmt::print("fmt   : [{:a}]\n", 3.0);
  fmt::print("fmt # : [{:#a}]\n", 3.0);
  fmt::print("fmt   : [{:a}]\n", 2.0);
  fmt::print("fmt # : [{:#a}]\n", 2.0);
  std::printf("std   : [%s]\n", std::format("{:a}", 3.0).c_str());
}

Output:
fmt : [0x1.8p+1]
fmt # : [0x1.8p+1] // 0x present with and without #
fmt : [0x1p+1]
fmt # : [0x1.p+1] // # forces a decimal point, not the prefix
std : [1.8p+1] // std::format omits the prefix

So:

• The 0x prefix is always emitted, like printf's %a (and unlike std::format).
• The # flag forces a decimal point, not the prefix.

Fix

Update the a row to use the example 0x1.8p+1, note that the 0x prefix is always emitted, and correct the description of #. The implementation is correct; only the docs are updated to match it.

Fixes #4657