I'll work through your first example and give you the rest for homework.
Step 1: Based on the original M/L (pre-scratches), determine what each horses' implied win probability is by using the following formula: Implied probability = 1 / (1 + odds). So, for the 3-1 for example, the implied probability is 1 / (1 + 3) = 0.25. Sorry if you already know this. Here are the implied probabilities based on your example
Odds Implied probability
1.6 0.38
3 0.25
4 0.20
6 0.14
10 0.09
15 0.06
Step 2: Take out the horse(s) who scratch. In this case, I took out your 6-5 shot
Odds Implied probability
3 0.25
4 0.20
6 0.14
10 0.09
15 0.06
Step 3: For each horse, take the horse's implied probability (e.g., 0.25 for the 3-1 shot) and divide it by the sum of implied probabilities for all horses. In this case, for the 3-1 shot you'd take 0.25 and divide by (0.25 + 0.20 + 0.14 + 0.09 + 0.06) and get 0.34. This represents the 'adjusted' probability after removing the scratch(es). Here are the 'adjusted' probabilities in your hypothetical example:
Odds Adjusted probability
3 0.34
4 0.27
6 0.19
10 0.12
15 0.08
Step 4: Finally, convert the adjusted probabilities back into odds by using the following formula:
Odds = (1-Probability) / Probability.
So, for the original 3-1, the adjusted odds would be (1-0.34) / 0.34 = 2-1
The full adjusted M/L therefore in your case 1 would be:
Original ML Adj. M/L
3.0 2.0
4.0 2.7
6.0 4.2
10.0 7.2
15.0 10.9
Case 2 would be the same methodology.
Case 3, same methodology as above and your intuition is correct. The impact will be lower. Think of it this way...if a longshot scratches, any given horse's chances of winning go up (and hence odds go down), but not by that much
Case 4: I wouldn't throw away completely. There's still going to be some information in the M/L even if the race switches surfaces. Though, not as much as before