Prevent absurdly low minsizes in yg-pe

[AdamISZ]

Fixes #889.
Prior to this commit, it was possible for the
yg-privacyenhanced.py script to go into an infinite
loop of attempting to halve the txfee setting, in case
the user selected (a) reloffer, (b) an absurdly low
minsize like 1 sat, and (c) a txfee of zero.
This occurred partially due to the randomization of minsizes
included in the yg-pe algorithm, which could lead to a
pathological zero value for minsize.
After this commit the minsize is automatically bumped to
the DUST_THRESHOLD value if it falls below it, preventing
the possibility of a nonsensical value.
Additionally, the loop which halves the txfee is no longer
allowed to continue beyond 20 iterations, as an extra sanity
check.

[AdamISZ]

Testing:

1. Reproduced the behaviour described in YG stuck at "Warning: too high txfee to be profitable, halfing it to: 0" #889 on master, using this on regtest:

python yg-privacyenhanced.py --datadir=. 88c3ed2aee24f69697d94463d2b116c1 -o reloffer -s 1 -t 0

(warning : if you try to do this test, be ready to find and kill -9 the process quickly as it eats up disk space fairly fast).

2. Confirmed that the behaviour is fixed by using this PR code and running the same command and seeing the expected message

2021-06-02 16:09:15,300 [WARNING]  Minsize was randomized to below dust; resetting to dust threshold.

followed by normal startup.

We could "gild the lily" by messing around with the exact value when it hits the limit, but using ridiculously low values around 10k sats is not something users should be doing anyway, since they don't actually want utxos that small.

[AdamISZ]

Also, it is probably worth mentioning that the edit to the loop (from while to for i in range(20) ) isn't logically necessary, but: this is the second time we encountered an edge condition leading to an infinite loop there, and since it's so nasty, it makes zero sense to not just prevent it.

[kristapsk]

ACK, tested before / after. Needs squash, of course. Note for other testers - you will not be able to reproduce this all the time, as it depends on randomization.