Optimise `predict_healthcareseeking` in `HealthSeekingBehaviour`

[tamuri]

Takes ~9% of runtime in mode 1. Did a quick test with using underlying numpy array. Seems to be much faster, but need to confirm numbers are the same.

            result = np.full(len(df), p[f'baseline_odds_of_healthcareseeking_{subgroup}'], dtype=float)
            age_years = df.age_years.values
            sex = df.sex.values
            region = df.region_of_residence.values
            li_urban = df.li_urban.values
            li_wealth = df.li_wealth.values

            # Predict behaviour due to the 'average symptom'
            if subgroup == 'children':
                result[age_years >= 5] *= p['odds_ratio_children_age_5to14']
            elif subgroup == 'adults':
                result[(age_years >= 35) & (age_years <60)] *= p['odds_ratio_adults_age_35to59']
                result[age_years >= 60] *= p['odds_ratio_adults_age_60plus']
            result[li_urban] *= p[f'odds_ratio_{subgroup}_setting_urban']
            result[sex == 'F'] *= p[f'odds_ratio_{subgroup}_sex_Female']
            result[region == 'Central'] *= p[f'odds_ratio_{subgroup}_region_Central']
            result[region == 'Southern'] *= p[f'odds_ratio_{subgroup}_region_Southern']
            result[(li_wealth == 4) | (li_wealth == 5)] *= p[f'odds_ratio_{subgroup}_wealth_higher']
            # Predict for symptom-specific odd ratios
            for symptom, odds in care_seeking_odds_ratios.items():
                symptom = df[f'sy_{symptom}'].values
                result[symptom > 0] *= odds
            result = (1 / (1 + 1 / result))
            result = pd.Series(result, index=df.index)
            # If a random number generator is supplied provide boolean outcomes, not probabilities
            if rng:
                outcome = rng.random_sample(len(result)) < result
                return outcome
            else:
                return result