Use numpy objects instead of pandas for core projection logic

[alexeybaran]

Use numpy objects instead of pandas for core projection logic. Example of modified init.py for CashValue_ME_EX1 attached

init.txt
(updated init.txt)

[fumitoh]

@alexeybaran the __init__.txt attached above doesn't include anything

[fumitoh]

CashValue_EX4 and the notebook are updated. Now 50% - 60% reduction in run time from CashValue_EX1.
has_surr_charge() is not used anymore in EX4, and now setting 'nan' to surr_charge_id() should serve the role of indicating no surrender charge. But this may be error prone, because when surr_charge_id is not found, then no surrender charge is assumed automatically. For example, when surr_charge_id() is type_3 but the type_3 column is missing in surr_charge_table by accident.

[alexeybaran]

Optimized lookup cells surr_charge_id() and premium_pp(t). It should half the time.
init.txt

[fumitoh]

This one is the same as the previous one.

[alexeybaran]

Sorry, I meant surr_charge_rate(t)

def surr_charge_rate(t):
    ind_row = np.minimum(duration(t), surr_charge_max_idx())
    return surr_charge_table.values.flat[surr_charge_table_column() + ind_row * len(surr_charge_table.columns)]

[fumitoh]

Updated EX4 and the notebook again.
Now EX4 runs in 20% to 30% of the time EX1 takes (70%-80% reduction)!