Interval->day patch

Please find attached a patch which adds a day field to the interval
struct so that we can treat INTERVAL '1 day' differently from
INTERVAL '24 hours' in DST-aware situations. It also includes a
function called interval_simplify() which takes an interval argument
and returns an interval where hours over 24 are promoted to days, e.g.,

template1=# select interval_simplify('3 months -11 days 79 hours 2
minutes'::interval);
     interval_simplify
--------------------------
3 mons -7 days -16:58:00
(1 row)

If anyone has better ideas for the name of this function, please let
me know.

I've modified the regression tests, but still need to add additional
tests for the interval_simplify function, and I want to add a few
more tests for the new interval behavior. Also, the docs will need to
be updated to mention the new behavior. I plan on doing this in over
the next couple of days.

This is some of the first C I've hacked, and the first patch I've
submitted that's more than a documentation or a simple one-liner (and
even that one got worked over pretty good :) ), so I fully expect
some mistakes to be found. Please let me know and I'll do my best to
fix them.

In timestamp.c, I suspect that AdjustIntervalForTypmod,
interval_scale will need some modifications, though I'm not quite
sure what this code is doing. I've left them as-is. I've made some
changes to interval2tm, but believe that the changes I've made may
not be adequate. Given sufficient instruction, I'll be happy to make
the necessary changes to these functions.

A few things I noticed while I was working:

In interval_mul and interval_div, I'm wondering whether 30.0 and 24.0
shouldn't be substituted for 30 and 24 in the non-integer-timestamp
code path, as these are floats. Perhaps it doesn't make a difference
for multiplication, but I see similar usage in interval_cmp_interval.
I've left the code as-is.

In the deconstruct_array calls in interval_accum and interval_avg,
the size of interval is passed as a magic number (16). I think this
could be abstracted out, such as #define SIZEOF_INTERVAL 16 to make
the code a bit more robust (albeit just a little). Is this a
reasonable change?

Michael Glaesemann
grzm myrealbox com