makeqstrdata: use an extremely accurate dictionary heuristic

[jepler]

Try to accurately measure the costs of including a word in the dictionary vs the gains from using it in messages.

This saves about 160 bytes on trinket_m0 ja, the fullest translation for that board before and after this change. Other translations on the same board all have savings, ranging from 24 to 228 bytes. The list below sorted the fullest translations (before the change) to the top. The numbers shown are the "bytes free in flash firmware" reported during a local build.

Translation     Before  After   Savings
ja              1164    1324    160
de_DE           1260    1396    136
fr              1424    1652    228
zh_Latn_pinyin  1448    1520    72
pt_BR           1584    1736    152
pl              1592    1640    48
es              1724    1816    92
ko              1724    1816    92
fil             1764    1800    36
it_IT           1896    2040    144
nl              1956    2136    180
ID              2072    2180    108
cs              2124    2148    24
sv              2340    2448    108
en_x_pirate     2644    2740    96
en_GB           2652    2752    100
el              2656    2768    112
en_US           2656    2768    112
hi              2656    2768    112

Deltas for adafruit_proxlight_trinkey_m0 in CI:

ja              1156    1292    136
de_DE           1280    1388    108
zh_Latn_pinyin  1472    1560    ...
fr              1508    1680
pt_BR           1604    1724
pl              1608    1648
es              1740    1820
fil             1764    1800
it_IT           1912    2044
nl              1940    2132
ID              2080    2204
sv              2340    2248
en_GB           2640    2740
en_x_pirate     2652    2748
en_US           2664    2748

[jepler]

The above stats apply to the FIRST commit only. The second commit may affect things a little bit, but only for the better. The problem with potentially accepting "negative valued" dictionary words only came into play with another change I'm testing locally which allowed for more dictionary terms.

[jepler]

I grabbed the stats across all boards. Here are the savings for all the builds that were within 200 bytes of full before this PR:

Board	Translation	Before	After	Savings
sensebox_mcu	ja	4	160	156
sensebox_mcu	de_DE	100	232	132
arduino_nano_33_iot	ja	112	268	156
feather_m0_rfm69	ja	144	324	180
blm_badge	ja	168	280	112
blm_badge	de_DE	172	272	100
arduino_mkr1300	ja	192	352	160

Some boards were a slight loss (why?) but none of them are seriously full:

Board	Translation	Before	After	Savings
makerdiary_nrf52840_m2_devkit	ID	300712	300672	-40
makerdiary_nrf52840_m2_devkit	pl	299408	299372	-36
pyboard_v11	pl	450184	450148	-36
same54_xplained	pl	553800	553764	-36
stm32f4_discovery	pl	458544	458512	-32
feather_stm32f405_express	pl	497164	497136	-28
pygamer	pl	14216	14192	-24
pybadge	pl	14628	14604	-24
winterbloom_big_honking_button	fil	49000	48976	-24
makerdiary_m60_keyboard	pl	301296	301280	-16
stm32f412zg_discovery	pl	477960	477948	-12
grandcentral_m4_express	pl	544380	544372	-8
feather_m4_can	pl	22308	22304	-4
makerdiary_m60_keyboard	ID	302604	302600	-4

Overall graph of the net change by board, sorted by savings:

[tyomitch]

The improvement in the heuristic's accuracy is excellent, I just wish the comment describing the new heuristic were more accurate.