**Update**: the most fundamental hole in the reasoning is after applying [Bézout's identity][1], which implies existence of $d$ and $k$ with $ed+\phi(pq)k=1$. Then, the sentence _where $d$ appears as the multiplicative inverse of $e$_ attempts to link the $d$ thus exhibited to the $d$ used in RSA. But why would these $d$ share more than their name, especially since the $d$ and $k$ exhibited by Bézout's identity are not unique, and (at least the usual form of) Bézout's identity does not state a relation between these multiple solutions?

We want either a more precise statement of Bézout's identity, or getting rid of it altogether. That's easy: start from the definition of $d$ in RSA (whatever that is), and prove that a suitable $k$ must exist, using fact 3 below.
___

The [proof as it stood at time of starting this answer][2] attempted to prove something that did not hold with the hypothesis that it then stated, which did not include
$$\;p\ne q\;\text{ or }\;\gcd(m,pq)=1\;$$
This is required in RSA (illustration: try $p=q=5$, $\phi(pq)=20$, $e=3$, $d=7$; encryption of $m=10$ followed by decryption yields $0$ rather than $10$ ).

Now $p\ne q$ is made explicit, satisfying said requirement. But it is not apparent where this is used.

Not coincidentally, the proof still has a serious gap at the point where $1^k$ appears, which implicitly uses that  $m^{\phi(pq)}\equiv1\pmod{pq}$, because:

- This proposition is wrong for some $m$, including $m=2q$ .
- [Fermat's little theorem][3] is invoked as a justification, but an hypothesis in FLT is that the modulus is prime, while $pq$ is not.
- FLT makes no mention of $\phi$ .

Useful standard facts (for all variables in $\mathbb Z$ unless otherwise noted):

1. If $p$ and $q$ are coprime, then $pq$ divides $x$ if and only if both $p$ and $q$ divide $x$ .
2. If $p$ and $q$ are distinct primes, then $p$ and $q$ are coprime.
3. The definition of $u\equiv v\pmod w$ is that $w$ divide $v-u$ ; or equivalently that there exists $k$ such that $u+kw=v$.
4. For $w>0$, the definition of $u=v\bmod w$ used in RSA encryption and decryption is that $u\equiv v\pmod w$ and $0\le u<w$ .
5. FLT: if $p$ is prime, then $y^p\equiv y\pmod p$ .

That allows deriving the helpful:

- if $p$ and $q$ are distinct primes, and both $p-1$ and $q-1$ divide $j-1$, and $j>1$, then $y^j\equiv y\pmod{pq}$ .

Proof hint: use fact 1 with $x=y^j-y$ , and other above facts.

___
Independently: it is used, but not stated, that the definition of RSA considered uses $d$ such that $ed\equiv1\pmod{\phi(pq)}$ . Another popular definition uses $ed\equiv1\pmod{\lambda(pq)}$ , where $\lambda$ is the [Carmichael function][4]. This definition is used in PKCS#1 and FIPS 186-4. It is mathematically satisfying, for it is necessary and sufficient, when $ed\equiv1\pmod{\phi(pq)}$ is merely sufficient.

Also, the proof would be clearer if it was restated a number of other things:

- $p$ and $q$ are primes;
- $N=pq$ ;
- whatever hypothesis on $m$ (commonly, that is $0\le m<N$, with further restriction to $\gcd(m,N)=1$ if the condition $p\ne q$ is not used);
- use of textbook RSA encryption $m\to c=m^e\bmod N$ ;
- use of textbook RSA decryption $c\to m'=c^d\bmod N$ (with a distinct notation for the original message and deciphered messages);
- what's to be demonstrated (this answer assumes that it is $m'=m$ for all $m$ satisfying the hypothesis made on $m$ ).

Also: there's a missing bit of reasoning, going from $m'\equiv m\pmod N$ to $m'=m$ .

Finally: textbook RSA is not a secure encryption algorithm (assume encryption of the name of someone in the class roll, which will be interrogated tomorrow; one can easily determine from the ciphertext and public key if that's her/him, or even who this is if the class roll is public).


  [1]: http://mathworld.wolfram.com/BezoutsIdentity.html
  [2]: https://crypto.stackexchange.com/revisions/51284/4
  [3]: http://mathworld.wolfram.com/FermatsLittleTheorem.html
  [4]: https://en.wikipedia.org/wiki/Carmichael_function