This will work with file of any size since it read and write the data lazily, however it may fail in case the number of columns explodes(which is unlikely). You can even check on your local machine by downloading the file.

splitrowvalue.py

import pandas as pd


chunksize = 1000
ip_filename = '/home/wolfram/Downloads/DataParserData.csv'
op_filename = 'datafile.csv'
prefix = 'cleansedquery_'

all_keys = set()
first_write = True

def update_values(key_value, all_keys=all_keys):
    missing_keys = all_keys - set(key_value.keys()) 
    key_value.update({key:0 for key in missing_keys})
    return key_value

for df in pd.read_csv(ip_filename, chunksize=chunksize, usecols=['CleansedQuery']):
    df['keys'] = df.CleansedQuery.apply(lambda x: [y.split(':')[0] for y in x.split(' ')])
    all_keys.update({value for row in df['keys'].values for value in row})
for df in pd.read_csv(ip_filename, chunksize=chunksize):
    df['key_value'] = df.CleansedQuery.apply(lambda x: {y.split(':')[0]: int(y.split(':')[1]) for y in x.split(' ')})
    df['key_value'] = df['key_value'].apply(update_values)
    for key in all_keys:
        df[prefix+key] = df['key_value'].apply(lambda x: x[key])
    
    df.drop('key_value', inplace=True, axis=1)
#     df.sort_index(axis=1, inplace=True)
    if first_write:
        df.to_csv(open(op_filename, 'w'))
        columns = df.columns
        first_write = False
    else:
        df.loc[: columns].to_csv(open(op_filename, 'a'), header=False)