The easier question first:
Currently, the smallest unit of account in the Bitcoin system is a satoshi (8 decimal places, 0.00000001 BTC). The total money supply is sub-divisible to 2.1 quadrillion (10^15), which is somewhat arbitrary (one nice property is it does fit as an IEEE double-precision floating-point number, which isn't used by the core network, but does make program interaction easier). If you run this against current global M1, that comes out to 2 cents/satoshi. It's basically a non-problem, although it seems to come up frequently. (If necessary, the protocol can be extended to subdivide to more decimal points. There aren't intrinsic problems - bitcoin values are stored as 64-bit signed ints by clients, which is good to 9.2 quintillion (10^18)).
Most people have switched to mBTC (milli) for pricing and eventually people will switch to uBTC (micro) all the way down to satoshis if it comes to it, so again, practical conversion to fiat isn't a problem. If 1BTC = $50K, then if you want to carry around $100 then you'd carry 2 mBTC or 2000 uBTC. Payment processors and wallets all do automatic conversion, so you may just end up pricing everything in local fiat even if your transactions are in Bitcoin.
The "deflationary" aspect is the trickiest to answer. There's a good summary page on the Bitcoin wiki on the money supply. What may be a surprise to some is that Bitcoin money supply is currently inflationary (11.11% this year) and in fact will be mildly inflationary for the next decade or so before it gets to <1%. The BTC/block rate halves about every 4 years and the growth rate goes down asymptotically until the last coin is mined out in 2140. Between now and then everyone (or the majority, at least) could agree to extend/alter the blockchain, or probably more likely, people could abandon Bitcoin for transactional use.
I used to think the deflationary aspect was a big deal - it encourages hoarding, slows spending/money velocity, and it'd be a real bummer to pay back loans denominated in a deflationary currency... but now I'm not so worried by it.
In the short term, the S-curve for Bitcoin adoption for transactional processing is just ramping up. It makes sense purely based on utility value of saved transaction fees. Payment processors provide instant fiat conversion so merchants and consumers suffer basically no currency risk. And while the current money supply is currently inflationary, the price of Bitcoin itself is mostly being driven (up) by conversion/transfer of wealth from existing fiat currencies into the Bitcoin network (aka speculation).
Since Bitcoin supply is algorithmically controlled (instead of arbitrary control by a centralized bank), it actually works well as a store of value. This of course complicates both the "medium of exchange" and "standard of deferred payment" aspects.
Personally, I think after the intial rush and *coin infrastructure get built out, other options, like Freicoin (demurrage), Peercoin (proof of stake, inflation), or even Dogecoin (wow, much coin) may all be better day-to-day transactional currencies. A lot of people are putting work into all kinds of crypto-coin exchanges, so I'd expect switching to different currencies to be pretty painless and maybe common/automatic.
Now, in an ideal world, a perfect currency would grow at exactly the right rate to provide optimal money velocity against actual economic activity (productivity, trading, energy, etc), but sadly in reality today, there's no part of that which wouldn't be gamed for individual gain (and at cost to the system/other participants). It'd be interesting if there was some sort of intrinsic algorithm for scaling, but as we can see by how some of the Chinese exchanges have been faking trade volume, that's also really tricky thing. This is actually one of the things on my list to do research on, but I haven't stumbled on anything too interesting yet that's tackling any of this.