Read day of month, month and year as prose

[austek]

Description of your changes

Follows #717 and #719, and is stacked on both — this branch contains those commits too, so please merge #719 and #717 first and this will reduce to its own two commits. Together the three cover the concrete cases reported on #3.

Three reported descriptions, all still reproducing on master:

reported in	expression	before	after
issue body (@francisdb, 2014)	0 0 0 1 1 ?	at 00:00 at 1 day at January month	at 00:00 on day 1 in January
issue body (@francisdb, 2014)	0/1 * * * * ?	every second from second 0	every second
comment (@Naxos84, 2017)	3/4 5/6 7/8 9/2 10/2 ? 2017/2	… every 2 months from month 10 every 2 years from year 2017	… every 2 months from October every 2 years from 2017

Three changes in descriptor/ produce that:

• Day of month, month and year get their own On phrasing instead of the generic "at <value> <unit>" template.
• describe(Every) drops the start clause when the period is 1 and the start is 0, since "every 1 unit from 0" is just "every unit". A meaningful start is kept — 5/1 * * * * ? still reads "every second from second 5", and 0 0/5 14 * * ? still reads "every 5 minutes from minute 0 at 14 hour".
• The unit word is omitted where the value already names it, so "from October" rather than "from month October". monthsInstance also no longer substitutes Object::toString when the expression is an Every, which is what produced "month 10" in the first place.

Day of week and the list forms follow, so the whole vocabulary agrees rather than only the three reported expressions:

field	single	list
day of month	on day 1	on days 1, 5 and 26
day of week	on Monday	on Monday, Wednesday and Friday
month	in March	in January, May and December
year	in 2005	in 2005 and 2006

A list reuses the bundle key of its single value form, passing the joined values as the argument, so a translator fixes one string rather than two. Only day of month needs a plural key of its own. Ranges and Every expressions are untouched — MON-FRI still reads "every day between Monday and Friday". The L and # day of week forms are untouched by this PR; they read "on the last Friday of the month" and "on the first Monday of the month" because #719 sits below it in the stack.

Impact Analysis

Output changes for expressions with an explicit day of month, day of week, month or year, whether single or a list, and for x/1 starting at 0. Callers asserting on those exact strings will need updating.

I diffed 50 Quartz and Unix expressions between this branch and the commit it stacks on: 21 lines change, all of them the cases above, nothing incidental. Seven pre-existing expectations in this repo are updated across the two commits here — three in CronDescriptorQuartzIntegrationTest, two rows of Issue126Test (the month clause of the #719 expression moves from "from month 1" to "from January"), one in CronDescriptorTest and one in ExecutionTimeQuartzIntegrationTest.

Two things reviewers should weigh:

• Day of week and list forms are included. Fixing months first left 0 10,44 14 ? 3 WED reading "… in March at Wednesday day", and fixing that in turn exposed lists still reading "at Monday, Wednesday and Friday days". Both are now consistent with the single value forms; see the table above. Hours and minutes keep their existing at 1, 2, 3 and 4 hours, since numeric units read fine as a list and are not part of Descriptions could be better #3.
• The five new keys are in the base bundle only, so the other seventeen locales fall back to English for these phrases — German reads "um 00:00 on day 1 in Januar". No locale throws; I checked all eighteen. This is the same fallback the bundles already rely on elsewhere (ro has no at today). Months and years take separate keys, in_month_x and in_year_x, even though English formats both as in {0}, precisely so a translator can decline them differently rather than being forced into one template. Translations need native speakers and I did not want to guess them.

One rough edge left in place: an expression restricting both day of month and day of week repeats the preposition, so 30 4 1,15 * 5 reads "at 04:30 on days 1 and 15 on Friday". That is the field concatenation in CronDescriptor.describe, not the phrasing changed here, and it was equally repetitive before ("at 1 and 15 days at Friday day"). Joining the fields properly means deciding an order for the whole sentence, which is the larger change #126 also asks for.

I am aware of the 2018 discussion where the template approach was set aside in favour of neural translation. This is not an attempt to reopen that — it is the minimum needed to make the specific expressions in #3 grammatical, and it leaves the per-language phrasing decisions to the bundles. Happy to close it if you would rather keep that direction clear.

TestDescriptor is left disabled; its expectations describe a different output shape (comma separated clauses, ordinals, am/pm) and reaching them is a larger change than this.

Checklist

• I have performed a self-review of my own code
• My changes generate no new warnings
• Existing tests pass: 838 tests, 0 failures, 0 errors (802 on master, plus 9 from Name the position of an nth day of week #719, 8 from Describe explicit minute 0 when seconds are not implicit #717 and 19 added here; 57 pre-existing skips). Every commit in the stack passes the full suite on its own.