Bug#21922414 CAST OF TOO BIG HEX LITERAL TO BIGINT UNSIGNED: BAD RESULT AND NO WARNING Bug#11757477 ARITHMETIC ERROR WHEN DEALING WITH BIGINT

Problem 1:
Item_hex_string::val_int() silently ignored any bytes except the last 8.
Fix: raise truncation warning if ignored bytes are non-zero.

Problem 2:
Hex_literal + int
was considered as binary + int in the sense that the result type was
DOUBLE.
The manual says that in numeric context, hex literal is 64-bit int;
reading the ctor also adds that it's then unsigned (unsigned_flag=1).
Fix: give it an implementation of numeric_context_result_type().
Consequence on parser_precedence.test: some calculations done
as DOUBLE are now done as BIGINT UNSIGNED so they don't permit
a substraction with negative result. Fixed by changing values;
intention of the test is preserved.