Warnings with clang when using long double for Lua floats.

In 'lcode.c', when adding constants to the list of constants of a
function, integers represent themselves in the cache and floats
with integral values get a small delta to avoid collision with
integers. (This change avoids creating artificial addresses; the old
implementation converted integers to pointers to index the cache.)

Some places don't need the "fast path" macro tointegerns, either
because speed is not essential (lcode.c) or because the value is not
supposed to be an integer already (luaV_equalobj and luaG_tointerror).
Moreover, luaV_equalobj should always use F2Ieq, even if Lua is
compiled to "round to floor".

By producing absolute line information at regular intervals, a simple
division can compute the correct entry for a given instruction.

The function 'changedline' tries harder to avoid calling
'luaG_getfuncline' plus small changes in the use of 'L->oldpc'.

Functions to set values in a table (luaH_set, luaH_newkey, etc.) receive
the new value, instead of returning a slot where to put the value.

Names in the parser and other details that do not change actual code.

- small corrections in the manual
- ldo.c: 'docall' -> 'ccall' ('docall' already used in 'lua.c')
- comments

This function was computing invalid instruction addresses when the
expression was not a multi-return instruction. (Virtually all machines
don't raise errors when computing an invalid address, as long as the
address is not accessed, but this computation is undefined behavior in
ISO C.)

Several details in code (e.g., moving a variable to the most inner
scope that encloses its uses), comments, parameter names, extra tests.