Improve missing value handling logic of ptable_heatmap

Our ptable_heatmap has suboptimal missing value handling logic, to close #230:

• Currently Non-PTableData typed data would be converted, but with the missing_strategy hard-coded as mean, meaning we're artificially creating non-existent data by default also user cannot control this behaviour, perhaps showing missing value as is (-) would be better.
• Is there a better way to resolve this than simply adding one more argument (on_empty sort of overlaps, perhaps we could partly reuse it)
• Import performance check show current import time, also round to int (should suffice as import time is on the 1000 level)

TODO list:

• Make sure a warning would be emitted (only) when data contains NaN/None
• Test missing value handling for both PTableData and others (pd.DataFrame) when containing missing values (NaN/None)