Normative: Fix spec bugs in numberformat.html caused by Unified NumberFormat

[sffc]

I changed part of the ToRawPrecision algorithm in ES2019 as part of Unified NumberFormat. However, in my work on NFv3, I realized that one change I made is unsound. This is in a crucial function, ToRawPrecision, which is used for all significant digits rounding.

Problem explanation: In ToRawPrecision, the e value is supposed to represent the power of 10 of the resolved number after rounding. However, my change in ES2020 computed e based on the input value before rounding.

Example values of x, p, n, and e, with xFinal for reference:

Formula: x ≈ n × 10^e–p+1

x	p	n	e	xFinal
1.11	3	111	0	1.11
11.1	3	111	1	11.1
1.11	2	11	0	1.1
11.1	2	11	1	11
9.999	3	100	1	10.0
9.999	2	10	1	10
9.9	3	990	0	9.90
9.9	2	99	0	9.9

Practically, the ES2020 algorithm works in all cases except those where the number rounds up to the next magnitude. A spec-compliant ES2020 implementation would need to return "9.99" instead of "10.0" in the fifth row.

Please check my logic!

All browsers still implement it correctly, but I would like to restore the old algorithm as soon as possible. I consider this an editorial change because it is restoring correct behavior after a mess-up in ES2020. (2021-05-06: This is still normative.)

Can we sneak this into ES2021? (2021-05-06: No, don't push this into ES 2021.)

[sffc]

I found another spec bug originating from Unified NumberFormat, which I added to this PR as 565a054.

[ryzokuken]

The PartitionNumberPattern revert looks great, the ToRawPrecision change is a bit more complicated but since it is exactly the same as the previous line, it should be all good.

[sffc]

2021-06-03: TG2 consensus: https://github.com/tc39/ecma402/blob/master/meetings/notes-2021-06-03.md#normative-fix-spec-bugs-in-numberformathtml-caused-by-unified-numberformat

[ryzokuken]

@gibson042 @leobalter could we have another review on this?

[sffc]

I would like to have this checked in. Can I have another review please?

[ryzokuken]

2021-05-25: TG1 consensus at https://github.com/tc39/notes/blob/master/meetings/2021-05/may-25.md#ecma402-status-update.

[ryzokuken]

@sffc do you think this is ready to merge?

[sffc]

Yes, if you all approve, this can be merged. Is there anything that needs to be done to fix the CI failure, or is that just flaky?

[ljharb]

It's unlikely to be flaky; if you rebase it i suspect you'll see a real lint failure.

[sffc]

I merged in the latest master, and CI is green now. 😃

[ryzokuken]

Okay, thanks, will all the boxes checked, let's merge this!