Next PR removing rules which ,make sense for English, but should not be
present in the Polish language files.

So we can run tests parallel with cmake, reducing test time and IO
overhead

Fixes: #1823, #1824, #1825, #1826, #1827

- Add crash test and vectors provided by @SEU-SSL
- Disallow dummy/null voice load (that causes incorrect translator
initialization)
- Fix empty `phondata` file load (that causes unitialized memory access)
- Limit max word length for RemoveEnding (causes buffer overflow)
- Limit punctlist initialization from embedded commands (buffer
overflow)
- Fix unitialized pitch in wavegen (DBZ and indexing problems)
- Properly zeroize stack variables before use in TranslateClause and
SetWordStress

TODO (in nextup PR): add & fix more vectors from fuzzer.

This PR removes two unnecessary rules in Polish. They both make sense
for English (probably someone just copied them from the English file),
but make pronunciation of several Polish words incorrect.

Without this function declaration, the build fails on modern C compilers
(ISO C99). This affects e.g. builds on Apple Silicon. This fix is also
needed in order to compile
[rhasspy/piper](https://github.com/rhasspy/piper) on Apple Silicon.

Build error was:
```
/Users/leo/Development/piper/build/p/src/piper_phonemize_external-build/e/src/espeak_ng_external/src/libespeak-ng/speech.c:872:3: error: call to undeclared function 'TranslateClauseWithTerminator'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
  TranslateClauseWithTerminator(translator, NULL, NULL, terminator);
  ^
1 error generated.
```

Without this function declaration, the build fails on modern C compilers (ISO C99). This affects e.g. builds on Apple Silicon. This fix is also needed in order to compile rasspy/piper on Apple silicon

This PR fixes the pronunciation of "bateau" in English, per
[Wiktionary](https://en.wiktionary.org/wiki/bateau).

When looking for dotted abbreviations we must limit ourself to the size
of the buffer used for translating the abbreviation.

When looking for dotted abbreviations we must limit ourself to the size of
the buffer used for translating the abbreviation.

This PR fixes pronunciation of the name "Shea" (pronounced like "Shay").

Fix pronunciation of
[Leschi](https://en.wikipedia.org/wiki/Leschi,_Seattle) ([Wikipedia
IPA](https://en.wikipedia.org/wiki/Leschi_(Nisqually))).

As reported in <https://bugs.freebsd.org/271486>, when the C++ standard
library provides std::sample, it causes a compile error when compiling
`speechWaveGenerator.cpp`:

```
  src/speechPlayer/src/speechWaveGenerator.cpp:197:56: error: reference to 'sample' is ambiguous
          unsigned int generate(const unsigned int sampleCount, sample* sampleBuf) {
                                                                ^
  src/speechPlayer/src/sample.h:23:3: note: candidate found by name lookup is 'sample'
  } sample;
    ^
  /usr/include/c++/v1/__algorithm/sample.h:95:17: note: candidate found by name lookup is 'std::sample'
  _SampleIterator sample(_PopulationIterator __first,
                  ^
```

Since `speechWaveGenerator.cpp` puts `using namespace std;` at the top,
fix the error by qualifying `sample` with an explicit `::`.

As reported in <https://bugs.freebsd.org/271486>, when the C++ standard
library provides std::sample, it causes a compile error when compiling
speechWaveGenerator.cpp:

  src/speechPlayer/src/speechWaveGenerator.cpp:197:56: error: reference to 'sample' is ambiguous
          unsigned int generate(const unsigned int sampleCount, sample* sampleBuf) {
                                                                ^
  src/speechPlayer/src/sample.h:23:3: note: candidate found by name lookup is 'sample'
  } sample;
    ^
  /usr/include/c++/v1/__algorithm/sample.h:95:17: note: candidate found by name lookup is 'std::sample'
  _SampleIterator sample(_PopulationIterator __first,
                  ^

Since speechWaveGenerator.cpp puts "using namespace std;" at the top,
fix the error by qualifying "sample" with an explicit "::".

Correct pronunciation of the word "banknote"
([Wiktionary](https://en.wiktionary.org/wiki/banknote)).

Reported by @kirill-jjj.

Before:
```
✗ espeak-ng -x -vru кошKa
k'oS(en)'eI(ru)
✗ espeak-ng -x -vru юMoney
'ju(en)w'0ni(ru)
✗ espeak-ng -x -vru ЮMoney                        
'ju(en)w'0ni(ru)
```

After:
```
✗ espeak-ng -x -vru кошKa 
k'oS(en)k'A:(ru)
✗ espeak-ng -x -vru юMoney
'ju(en)m'Vni(ru)
✗ espeak-ng -x -vru ЮMoney
'ju(en)m'Vni(ru)
```

I have not noticed any negative side-effects and tests are still
passing.
But I think there should be a more precise testing, because translator
code is very complex (and poorly organised, and not documented, and...)

OK, now CI is green again. It is github-CI related issue, I think.

The root cause is that libgcc-s1:i386 fails to be installed as
dependency (for unknown reason), so it is now installed explicitly.

Next bunch of fixes, including unused, non-consts and statics hunting.
Also improves cmake-based installation on MacOS (default data path and
rpath).