refactor: improve performance for hand calculations further

[Nihisil]

Closes #171

@Apricot-S fixed hand divider, so now it produces empty output if hand is not agari.

With that improvement we can simplify hand calculation and remove additional is_agari check from there.

All 2.1kk validation set hands pass with this change.

Benchmark results

Benchmark was run with n=5 and limit=200000.

Old

• Throughput: 31165 hands/sec (based on median)
• Avg per hand: 0.032ms (based on median)

New

• Throughput: 37634 hands/sec (based on median)
• Avg per hand: 0.027ms (based on median)

Overall profile

Metric	Old	New	Change
Total profile time	13.870s	12.384s	-10.7%
Total profile calls	56,573,351	54,714,452	-3.3%

estimate_hand_value function

Metric	Old	New	Change
Own time	2.404s	2.274s	-5.4%
Cumulative time	13.828s	12.347s	-10.7%
Callee calls	10,221,010	10,017,929	-2.0%
Callee tottime	4.484s	3.847s	-14.2%
Callee cumtime	11.424s	10.073s	-11.8%

[Apricot-S]

No breaking changes are required if we do the following:

hand_options = HandDivider.divide_hand(tiles_34, melds)
is_kokushi = not is_open_hand and config.yaku.kokushi.is_condition_met(None, tiles_34)

if not hand_options and not is_kokushi:
    return HandResponse(error=HandCalculator.ERR_HAND_NOT_WINNING)

...

if not calculated_hands:
    return HandResponse(error=HandCalculator.ERR_HAND_NOT_CORRECT)

[Nihisil]

Do you think it is worth to keep comparability here? It was really rare case and hack with additional error response. Now we have a chance to fix it properly

[Apricot-S]

Understood.