power: qpnp-fg-gen3: Fix possible integer overflow

Currently, DIV_ROUND_CLOSEST is used while finding the delta
between initial and final cc_soc_sw in terms of percentage. This
can go beyond 32 bits with the intermediate multiplication before
it is down-converted back to less than 32 bits. Fix this.

Also, when the cc_soc_sw delta is less than 50 %, discard the
data so that the last learned capacity won't be updated.

While at it, fix a similar use of DIV_ROUND_CLOSEST in another
place fg_encode_default() where an integer overflow can happen
if it is passed with large integer values.

Change-Id: Ia2b0f5456663a90bcd32a4fed52d1e579cff4e1f
Signed-off-by: Subbaraman Narayanamurthy <subbaram@codeaurora.org>