Given the placement of the chronograph close to the muzzle, the decrease in velocity over time, and the change in SD there are a few options, but no meaningful conclusions can be made without doing another test under more controlled conditions.
1. The chronograph may have moved and angled a bit so that the bullet crosses the sensors at an angle. It's really measuring speed by measuring the time (by measuring the voltage that accumulates in a capacitor between the time the first and second sensor is tripped) it take the bullet to cross the fixed distance between the sensors. The problem is the distance changes if the unit is angled up or down or side to side. The distance between the sensors is really the hypotenuse of a right triangle, and the more the unit is angled relative to the path of the bullet, the larger the "short" side of the triangle gets, and the more the hypotenuse will grow relative to the baseline length between the sensors.
The two largest and most common mistakes I see with shooter using chronographs are a) failing to ensure the chronograph is both level and precisely aligned with the path of the bullet from muzzle to target, and b) failing to ensure the chronograph is mounted securely so that its alignment doesn't change. Most shooter set them up haphazardly, then blame the chronograph when they get bad data.
2. The velocity is consistent after your first 30 rounds, the extreme spread is narrowing as reflected by both the extreme spread numbers and the improving SD. More importantly, your velocity numbers in the last two sets are about what I would expect from 25 grains of BLC-2 in an 18" barrel with a 55 gr FMJ-BT.
3. You started with a "relatively new rifle" so some of the decrease might be due to the barrel breaking in. There's also the issue of heat. 90 rounds in 10 minutes isn't exactly a low rate of fire and the barrel will indeed get quite hot. Hot barrels expand very slightly, and as it expands the bore enlarges very slightly as well, perhaps enough to account for some of the velocity loss.
4. BLC-2 is temperature sensitive, but in the .223 and .308 that works out to something on the order of about 1 to 2 fps per degree F, so you'd have to show at least a 200 degree F change in temp to get the velocity differences you are seeing, and given the effects of the round heat soaking in a hot chamber, I'd expect the later rounds to be faster, the opposite of what you observed.
-----
My advice is to ensure your chronograph is properly aligned and mounted securely, so that you are shooting directly across the sensors. In other words, when you look down the rifle at the target the chronograph should be level, the Vs from the supports for the sensors should be aligned/parallel to each other, and you should be shooting directly over the middle of the each sensor. That's actually far more important than muzzle to chronograph distance. If you're using a powder with little smoke, and a small round with less muzzle blast, you can get by with a closer chronograph distance, provided the chronograph still doesn't wobble or change alignment when you shoot.
Once you've gotten things properly set up, repeat the test firing 1 shot out of a clean bore, ignoring that cold bore data point, and then firing one round per minute, recording 10 round groups. Shoot three groups and then compare the data for all 10 groups.
You'll have an increase in temp over the whole 30 rounds, but it should be much less and the first 10 shots should reflect more or less cool barrel performance, while the last two groups will show a more representative warm barrel performance. It will only take 31 minutes and you'l have better data than 90 rounds in 10 minutes, and you'll be abusing the barrel a lot less.
If you see a consistent decrease in velocity again, then it's probably an artifact of the barrel warming up, although the change in velocity should be a lot less than you saw before.
If you don't see a significant difference between the first, second and third groups, then your initial results were probably an artifact of breaking in the barrel.