gh-146640: Optimize int operations by mutating uniquely-referenced operands in place (JIT only)

[eendebakpt]

In this PR we add inplace binary operations for int type, similar to #146397. A complication compared to the float case is that the int operations can return small ints which should not be modified in place. In this PR we:

• Use PyJitRef_MakeUnique(sym_new_compact_int(ctx)); to mark that a symbol is an int that is either unique or one of the small ints.
• Add _BINARY_OP_ADD_INT_INPLACE (with variations for the other operations) that handles both small ints as input and overflows. An alternative to handling these cases in the opcode would be do deopt. Here we decided not to deopts because this could lead to many deopts.

Micro benchmark results:

Pattern	main (ns/iter)	branch (ns/iter)	Speedup	Notes
total += a*b + c (non-small)	34.4	27.8	1.24x	inplace mutation on unique a*b result
total += a + b (non-small)	19.8	13.0	1.52x	a+b result is unique, inplace on +=
t = a + b (plain assign)	13.4	14.1	--	no +=, not affected
total += a*b + c*d (chain)	40.4	34.6	1.17x	two unique products, inplace on +
total += a*b + 1 (small int)	23.2	24.1	--	TARGET is small int, fallback to standard

All values are non-small integers (> 1024) unless noted.

Selected pyperformance benchmarks:

### deltablue ###
Mean +- std dev: 1.95 ms +- 0.03 ms -> 1.97 ms +- 0.03 ms: 1.01x slower
Not significant

### scimark_fft ###
Mean +- std dev: 208 ms +- 5 ms -> 206 ms +- 1 ms: 1.01x faster
Not significant

### scimark_lu ###
Mean +- std dev: 58.1 ms +- 0.2 ms -> 58.7 ms +- 0.8 ms: 1.01x slower
Not significant

### scimark_monte_carlo ###
Mean +- std dev: 41.7 ms +- 1.7 ms -> 41.6 ms +- 0.8 ms: 1.00x faster
Not significant

### scimark_sor ###
Mean +- std dev: 66.8 ms +- 0.6 ms -> 64.3 ms +- 0.5 ms: 1.04x faster
Significant (t=26.88)

### scimark_sparse_mat_mult ###
Mean +- std dev: 3.64 ms +- 0.03 ms -> 3.64 ms +- 0.04 ms: 1.00x slower
Not significant

### spectral_norm ###
Mean +- std dev: 56.4 ms +- 0.2 ms -> 51.5 ms +- 0.4 ms: 1.10x faster
Significant (t=79.61)

• Issue: Optimize int operations by mutating uniquely-referenced operands in place (JIT only) #146640