Remove a false performance claim in addRowMultiple, and the dead macros behind it - #77

The restrict local in DenseRow::addRowMultiple, removed in the previous
commit, was the only place it was ever used.

All three definitions go, not just the one this build takes.  The macro is
defined once per compiler branch -- __restrict for MSVC, __restrict for
GCC/clang, and empty for the fallback -- and removing only one branch would
leave a macro that exists on some compilers and not others, which is worse
than either keeping or removing it.

stdinc.h is installed by both build systems, so this is nominally an API
removal, as PR #60's FlattenNamespace change was.  There is no ABI effect: a
macro emits no symbol.  Nothing outside the project has reason to use a
MATHICGB_-prefixed compiler-portability macro, and anything that did would
already have had to cope with it expanding to nothing on unrecognized
compilers.

Verified: 246/246 tests pass.

Several neighbouring macros in the same blocks look equally unused, and some
are visibly broken -- __attribute__(pure) is missing its inner parentheses and
would not compile if anything expanded it.  Auditing those is a separate pass.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

MATHICGB_RESTRICT, removed in the previous commit, turned out not to be alone.
Auditing its neighbours in the same three compiler blocks found five more with
no uses anywhere in the project:

  MATHICGB_ASSUME_AND_MAY_EVALUATE
  MATHICGB_MUST_CHECK_RETURN_VALUE
  MATHICGB_NOTHROW
  MATHICGB_PURE
  MATHICGB_RETURN_NO_ALIAS

What settles it is that in the GCC/clang branch none of the five would
compile if anything did use them.  Four are written __attribute__(x) where
the attribute syntax needs __attribute__((x)):

  #define MATHICGB_RETURN_NO_ALIAS         __attribute__(malloc)
  #define MATHICGB_NOTHROW                 __attribute__(nothrow)
  #define MATHICGB_PURE                    __attribute__(pure)
  #define MATHICGB_MUST_CHECK_RETURN_VALUE __attribute__(warn_unused_result)

and the fifth has its while(0) inside the do block rather than closing it.
Checked by expanding each one in a one-line translation unit against GCC
11.4: MATHICGB_PURE gives "error: expected '(' before 'pure'", its three
siblings give the same shape, and MATHICGB_ASSUME_AND_MAY_EVALUATE gives
"error: expected primary-expression before '}' token".  So the first use of
any of them, on the compiler every current build uses, is a build failure.
They have been that way since 2013 and nothing noticed, because nothing uses
them.  A facility nobody can adopt is not worth keeping; adding one back
correctly is two lines.

The MSVC spellings look right and the fallback branch defines them empty, but
all three copies go together -- leaving a macro that exists on some compilers
and not others is worse than either keeping or removing it.  The five ///
comments in the MSVC block go too, since they document macros that no longer
exist.

stdinc.h is installed, so this is nominally an API removal, with no ABI
effect.  Macaulay2 is the only known consumer and uses none of them: grepping
its tree at 169ac29181 for all six macro names returns nothing outside our own
submodule, and the only MATHICGB_* identifiers it references anywhere are
MATHICGB_LIBRARIES, MATHICGB_INCLUDE_DIR, MATHICGB_FOUND, MATHICGB_VERSION_
STRING, MATHICGB_DEBUG and MATHICGB_NO_TBB -- things it sets or probes for,
not attributes it consumes.  It also cannot reach these definitions: it
includes only mathicgb.h and mathicgb/mtbb.hpp, neither of which includes
stdinc.h.

Two near neighbours are deliberately left alone, because the obvious grep
calls them dead and they are not.  MATHICGB_ASSUME has no uses outside
stdinc.h but is the non-debug definition of MATHICGB_ASSERT further down the
file, so removing it would break all 898 assertions in release builds -- it
survives immediately above the first hunk here, while the macro whose comment
begins "As MATHICGB_ASSUME, but..." does not, which is correct but reads
asymmetrically.  MATHICGB_CONCATENATE is used by
MATHICGB_CONCATENATE_AFTER_EXPANSION, which MATHICGB_UNIQUE needs.

Verified: 246/246 tests pass.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>