D5308: store: don't pass 'atomictemp=True' while appending to fncache

[Yuya Nishihara]

>   > As I said, we'll also need to somehow hide partial data from another
>   >  reader processes. There's no read lock.
>   
>   
>   Read locks! was there a previous attempt on implementing them? I will try to take a shot at implementing them. I am unable to find any good detail except mentions in wiki or mailing list archive.

Another option is to ignore trailing data after `\n`. That's how the revlog
parser gets around partially-written entry.

In either way, writing fncache non-atomically wouldn't be strictly compatible
with older hg clients. I think that's okayish, but I should note that.