Update to allow Alignment to run multithreaded

Modifications to allow running multithreaded:

• rewrote AlElementHistos to use Accumulator::Histogram
• changed from mutex to shared_mutex to profit from readonly locks
• create AlEquations object in each event call on the stack and add at end of event to minimize number of write locks
• AlignAlgorithm no longer derives from GaudiHistoAlg
• update gaudiiter to support running multi-threaded (works on MDF files only)

This addresses most of what is discussed in #30

This also fixes a crash in a test on clang that was due to a bug in the IO of the Al::Equations object.

@graven, @jcobbled

Alignment/AlignKernel/include/AlignKernel/AlEquations.h

@@ -30,7 +30,6 @@ namespace LHCb::Alignment {
     size_t           m_numHits{0};
 
   public:
-    OffDiagonalData() : m_matrix(), m_numTracks( 0 ), m_numHits( 0 ) {}
     Gaudi::Matrix6x6&       matrix() { return m_matrix; }
     const Gaudi::Matrix6x6& matrix() const { return m_matrix; }
     size_t                  numTracks() const { return m_numTracks; }
@@ -167,7 +166,7 @@ namespace LHCb::Alignment {
     /// object was computed around the same parameter set, and if not, correct for the difference.
     void add( const Equations&, bool correctDeltaAlpha = false );
 
-    std::ostream& fillStream( std::ostream& os ) const;
+    std::ostream& fillStream( std::ostream& os, int verbosity = 0 ) const;
 
     const ElementContainer& elements() const { return m_elements; }