linker.parameter_estimate_comparisons_chart() improvements

[samnlindsay]

linker.parameter_estimate_comparisons_chart() currently displays the m and u probabilities as log odds. This has its advantages, but the original chart was designed for probabilities (fixed x-axis between 0 and 1).

Need to make some visual tweaks to make the log odds version easier to interpret, and enable switching between probabilities and log odds (a log_odds=False argument or a button in the chart to toggle different formats).

Issues

• Parameters are estimated as probabilities with an effective tolerance of em_convergence. If this is set to 0.01 then we must expect the same m value to potentially be estimated as 0.0004 (log odds = -7.824) or 0.0104 (log odds = -4.5564). These values are consistent with one another given our em_convergence but on a log odds scale these small differences are amplified, implying that different training runs have estimated significantly different values.
• The probability scale is simple to interpret (left extreme = 0, right extreme=1) but the log odds scale is (potentially) $-\infty$ to $+\infty$. This means it's not immediately clear what range is being shown, and where the zero line is. (See example below)
• Related to the previous point, because of this moveable range, it's important to have the x-axis labels visible for each chart to be able to judge differences between points. (See the below)

Fixes

• MUST - Add zero line
• SHOULD - Allow probability version of chart (either with an argument or a toggle switch in the chart itself)
• COULD - Add axis labels to all facets for log odds version
• COULD - Add subtle lines extending from zero to the points (resembling match_weights_chart()) to highlight positive or negative log odds