[10.x] Improve decimal cast precision

[timacdonald]

Remove manual manipulation and require bcmath to use the decimal cast.

This should be noted in the 10.x upgrade guide.

[timacdonald]

On hold until #45492 is merged.

[driesvints]

@timacdonald I guess this can be continued to work on now that PR is merged?

[driesvints]

@timacdonald I moved the bc-math requirement to "suggest" which is what you wanted to do I think.

[barryvdh]

If it's in the suggest, then it crashes?
This is different then #45457 which checks if the extension is loaded. In this version for 10.x it always used bcmath

[driesvints]

@barryvdh it's in "suggest" because you don't have to use the decimal cast per se. It's suppose to crash if you don't have it installed. That's why it's a "suggest" and not a hard dependency.

Also, I think you wanted to link to a different PR?

[barryvdh]

If you don't want to require it, maybe better to keep the fallback? Maybe add an exception if the float precision is to low? Something like f66db46
(Removed the exception in #45580 for 9.x now)